Accelerating Computational Material Discovery and Learning: First Experiences with MatterSim

Accelerating Computational Material Discovery and Learning: First Experiences with MatterSim | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Accelerating Computational Material Discovery and Learning: First Experiences with MatterSim Gia Khanh Pham, Ulrick Edwin Nguegang Tsopmo, Marta Segura Cubells This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8143147/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Artificial Intelligence (AI) is reshaping materials science by accelerating discovery and enhancing education through data-driven approaches. This study explores the integration of Microsoft’s open-source tools, MatterGen and MatterSim, into academic and research workflows. MatterGen employs generative diffusion models to design novel inorganic materials, guided by property constraints and fine-tuned using large datasets. MatterSim, a machine-learning-based simulation engine, predicts thermodynamic, mechanical, and structural properties under realistic conditions with exceptional speed and accuracy. This work does not aim to propose a new ML model or benchmark, but to explore human-centric integration of MatterSim into materials education and early-stage research. In the first phase of this project, the focus was on evaluating MatterSim through simulations such as structural relaxation, phonon analysis, and molecular dynamics, supported by a custom-developed graphical user interface. The initial results showed good agreement between MatterSim predictions and reference data, confirming its reliability for educational and research purposes. Key benefits include a better understanding of material properties in a short time, improved accessibility for students and researchers to computational material discovery using AI, and reduced development time and costs. However, limitations such as hardware requirements, incomplete thermodynamic workflows, and dependence on high-quality training data were identified. Future directions involve exploring the second Microsoft’s tool, MatterGen, and combining generative and predictive models for rapid screening. This work underscores the transformative potential of AI in materials science and provides practical recommendations for its adoption in both academic and industrial contexts. The custom-developed graphical user interface, integrated with MatterSim and implemented in Python, is published on GitHub as open source ( https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git ). Artificial Intelligence Machine Learning MatterGen MatterSim Computational Materials Discovery Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1 Introduction Artificial Intelligence (AI) plays a game-changing role in materials science by offering added value for education and research & development (R&D). Its integration into academic and industrial contexts enables significant advancements in data-driven methodologies and personalized learning environments [ 1 , 2 ]. Key applications of AI in materials science education and general pedagogy include: Virtual Labs & Simulations: AI enables realistic modeling of complex phenomena like microstructure evolution and fatigue, enhancing hands-on learning [ 3 , 4 ]. Predictive Materials Design: Machine learning supports property prediction and accelerates the development of novel materials [ 1 , 5 ]. Automated Microstructure Analysis: AI tools classify phases and detect defects in material images, streamlining experimental evaluation [ 6 , 7 , 8 ]. Adaptive Learning Systems: Intelligent platforms personalize content based on learners’ progress and knowledge levels [ 9 , 10 , 11 ]. Generative AI for Teaching: AI assists in creating tailored exercises, assessments, and explanations for advanced topics [ 12 , 13 ]. AI tools in materials science education offer benefits in personalized learning, enabling students to progress at their own pace with tailored feedback. They enhance data literacy by providing hands-on experience with real-world datasets and bridge the gap between theory and practice through simulations without costly lab setups [ 14 ]. Students can solve problems faster by using AI for property prediction and process optimization, and they gain industry-relevant skills that prepare them for modern, data-driven R&D environments [ 15 ]. Beyond education, Artificial Intelligence (AI) has become increasingly prominent in the R&D of novel materials. Numerous case studies and projects have demonstrated its applicability across several key domains [ 16 , 17 ]. Machine learning techniques facilitate the prediction of material properties, including mechanical, thermal, and electrical characteristics [ 18 ]. In the filed of material discovery, Bayesian optimization and graph neural networks are employed to identify promising new compounds, while surrogate models significantly accelerate high‑throughput simulations [ 19 , 20 , 21 ]. Convolutional Neural Networks (CNNs) contribute to microstructure analysis [ 22 , 23 ], and generative models enable the inverse design of materials tailored to specific target properties [ 24 , 25 ]. Process optimization benefits from Bayesian and semi‑supervised learning approaches [ 26 ], whereas the development of sustainable materials leverages multi‑objective optimization to balance performance metrics [ 27 ]. Applying AI-driven methodologies in materials R&D has the potential to significantly shorten development cycles and foster the creation of novel material solutions [ 28 ]. Table 1 presents practical examples from industry and academia, demonstrating how artificial intelligence contributes measurable value and yields remarkable outcomes in materials R&D. ​ Table 1 Industry and Research Case Studies in AI-Driven Materials Development. Organization Goal AI Approach Result Google DeepMind Materials discovery (batteries, semiconductors, catalysis) GNoME (Graph Neural Networks for Materials Exploration) 2.2 million new crystals predicted; ~380,000 identified as stable; 736 already synthesized [ 29 ] Meta FAIR Large-scale materials datasets and surrogate models Open Materials 2024 (OMat24) dataset; Graph Neural Networks 118 million DFT calculations on 52 million structures; enables rapid property screening Academic Autonomous Lab Lawrence Berkeley National Lab (LBNL) – A-Lab Autonomous synthesis Self-driving lab: robotics + AI decision-making + ML-based characterization In 17 days, 41 new compounds synthesized from 58 targets; lab operates 50–100× faster than manual workflows [ 30 ]. Max Planck Institute for Iron Research (MPIE) + DeepMetis Mechanics simulation Deep-learning surrogate solver (DNN) 8,300 times faster than standard FEM solvers for elasto-plastic materials [ 31 ] Toyota Research Institute (TRI) & Partners (Stanford/MIT) Battery R&D Bayesian optimization Faster lifetime prediction [ 32 ] Citrine Informatics + HRL Laboratories (Boeing & GM) Design of 3D-printable alloys for aerospace. ML screening cycle-time reduction from years to days for candidate search & screening [ 33 ] Fraunhofer IWM Data-scarce scenarios Semi-supervised learning Lower annotation cost [ 34 ] Leibniz-IVW + DFKI + Fraunhofer ITWM (ML4SIM) Fiber composite manufacturing process ML-accelerated simulations (surrogates), anomaly detection Faster, robust process design [ 35 ] Forschungszentrum Jülich Perovskite PV – inverse design of organic hole transport materials High-throughput synthesis + ML for rapid optimization New molecules enhancing photovoltaic efficiency identified within weeks [ 36 ] A highly impressive example of AI tool development for R&D in new materials comes from Microsoft. Microsoft has developed two advanced AI tools, MatterSim and MatterGen, that are transforming materials science through rapid simulation and generative design [ 37 ]. MatterGen is a generative model for materials design [ 38 ]. It can create novel inorganic material structures from scratch, tailored to user-defined property targets (such as hardness, conductivity, or magnetism), using a prompt-driven approach. MatterSim is a deep learning model for simulating and predicting material properties, such as mechanical, electronic, and thermodynamic behavior, under real-world conditions across a wide range of temperatures and pressures [ 39 ]. It evaluates material candidates efficiently and with accuracy comparable to first-principles methods. Both MatterGen and MatterSim have been freely available since early 2024 as open-source projects and are accessible via Microsoft’s Azure Quantum Elements platform. These tools offer substantial benefits for both academic and industrial research [ 37 ]. In academic settings, they help accelerate prototyping by reducing the number of costly experiments, thanks to their ability to efficiently narrow down candidate materials. In industrial R&D, they enable fast screening and prioritization of materials for applications in manufacturing, energy, and electronics. For instance, researchers can use these tools to design and simulate new electrode compositions for battery materials. In the development of magnets and fuel cells, they allow for the optimization of structures and the prediction of performance before synthesis. For photovoltaic applications, MatterGen generates candidate materials, and MatterSim assesses their thermal and mechanical stability over typical PV operating temperatures (e.g., -40°C to + 85°C). However, optical properties and light-matter interactions (e.g., bandgap engineering, absorption spectra) require complementary electronic-structure calculations (DFT, GW, Bethe-Salpeter equation) or device-level simulations. Additionally, they support rapid calculation of phase diagrams, enabling the prediction of material stability and transformations under varying conditions. The novelty of this work lies in (i) implementing and open-sourcing a human-centric GUI for MatterSim tailored to teaching and capstone projects, (ii) critically assessing which simulation tasks are reliable enough for educational use, and (iii) documenting practical limitations and validation strategies against DFT and reference data in a realistic university environment. Our contribution is positioned at the interface between cutting-edge AI models and their practical adoption in educational and applied research contexts, with explicit documentation of usability, limitations, and best-practice workflows. This study explores the application of advanced AI tools, MatterSim and MatterGen, for research and education in materials science. [ 40 ]. The focus of the project lies in the evaluation and analysis of these tools, aiming to assess their practical suitability for materials development. While MatterSim and MatterGen are well documented and benchmarked by Microsoft, there remains a gap between these advanced tools and their practical integration into human-centric educational settings and project-based learning in materials science. Recent work on AI-enabled virtual labs and adaptive learning emphasizes the importance of such tools in education but also highlights an implementation gap in real teaching practice. Our aim is to: (i) bridge this gap by demonstrating how MatterSim can be integrated into a capstone project and course setting, (ii) provide an open-source GUI and workflow that lowers entry barriers for students without strong Python or command-line expertise, and (iii) document where current tools are robust enough for teaching and where their limitations require instructor guidance or complementary DFT validation. The primary objectives of the study include the installation and configuration of the tools, the execution of test simulations, and the assessment of simulation accuracy. In addition, aspects such as accessibility, usability, and computational efficiency are examined. An important task of the project is to create a graphical user interface (GUI) for convenient use and to document the entire workflow – from installation to application – in a way that enables other students and researchers to begin using these tools immediately, without significant setup time or technical barriers. While both tools are considered, the first phase of the project places particular emphasis on an in-depth analysis and practical application of MatterSim, due to its capabilities for simulating material properties under realistic conditions. The study aims to contribute to the broader understanding of how AI-driven simulation can enhance both academic learning and industrial R&D processes in materials science. 2 Theoretical Foundations of Microsoft’s AI Tools: MatterSim and MatterGen MatterGen uses a generative diffusion model which jointly predicts atomic structure, elements, and lattice parameters, fine-tunable for property constraints [ 38 ]. MatterGen is like a "creative AI scientist" that designs entirely new materials from scratch, using advanced generative algorithms. Traditional methods can only select or filter from materials that are already known; MatterGen, however, can imagine materials that have never existed, following a multi-step process [ 41 , 42 ]:​ Diffusion-Based Generation: The process starts with a random cloud of atoms in space. MatterGen uses its neural network to gradually arrange these atoms into a stable crystal structure, much like turning random static (noise) into a recognizable image.​ Property Conditioning: MatterGen can be given precise targets, such as high hardness, strong magnetism, or a specific chemical formula. The model continuously adjusts its output to meet these goals, generating materials tailored to specific requirements.​ Learning from Examples and Fine-Tuning: The tool was trained and is trained on a dataset of more than 600,000 real, experimentally confirmed crystal structures. For projects requiring niche or rare materials, a small labeled dataset can be provided, and MatterGen will adapt to suggest optimized candidates Several examples have been reported showcasing MatterGen’s capabilities in materials development [ 38 ]: To create a super-hard material for cutting tools, MatterGen generated over 100 candidates with a bulk modulus exceeding 400 GPa, significantly harder than most common metals. In comparison, classical screening methods identified only 40 such candidates in existing databases. Researchers used MatterGen to design a new material targeting a bulk modulus of 200 GPa. After synthesis and laboratory testing, the measured hardness was 169 GPa, demonstrating how closely the model’s predictions aligned with real-world results. MatterGen can generate strong magnetic compounds by setting “high magnetic density” of 0.2 Å −3 as a design goal. With just 180 density functional theory (DFT) property calculations, MatterGen is able to identify up to 18 stable, unique, and novel (SUN) structures exhibiting a magnetic density greater than 0.2 Å⁻³. The second AI tool from Microsoft, MatterSim, is a simulation engine powered by machine learning that predicts material behavior under real-world conditions with high speed and accuracy. It leverages deep graph neural networks , uncertainty-aware sampling, and active learning. MatterSim functions as a zero-shot machine learning force field (MLFF), enabling end-to-end structure-to-property predictions, and operates across a wide range of elements, temperatures, and pressures. The main functional capabilities and core advantages of MatterSim are summarized as follows [ 39 ]: Deep Learning Physics Engine: MatterSim learned from millions of physics-based calculations ("first-principles" data) covering thousands of different materials. It runs atomistic simulations: mimicking how a material’s atoms vibrate, move, and respond as conditions change.​ Wide Range Coverage: It can predict thermodynamic, mechanical, and structural properties across the periodic table, for conditions from cold (0 K) to extremely hot (5000 K), and low to very high pressure (up to 1000 GPa).​ Efficiency and Accuracy: MatterSim provides results comparable to advanced quantum-mechanical methods, but within seconds instead of hours or days.​ Representative use cases of MatterSim demonstrate its versatility and predictive power: Phase diagram computation: MatterSim can generate the phase diagram of a novel battery material within minutes, drastically reducing the time and experimental effort traditionally required. Prediction of mechanical properties under extreme conditions: The tool enables accurate estimation of the mechanical strength of superconductors across a broad temperature range from − 200°C to + 1200°C. Simulation of lattice behavior: MatterSim can model how the crystal lattice of a newly developed magnetic material responds to external stimuli such as intense magnetic fields or thermal stress. MatterGen and MatterSim are designed to work synergistically. MatterGen can be used to generate promising new material candidates based on specified target properties. These candidates can then be evaluated using MatterSim, which rapidly predicts their stability, manufacturability, and functional performance under realistic conditions. For instance, when developing a material for solar panels that requires high electrical conductivity and thermal stability, MatterGen can generate suitable candidates that meet these criteria. MatterSim assesses thermal and mechanical stability over typical PV operating temperatures; optical and light–matter effects are handled by separate methods (DFT, GW, Bethe–Salpeter, device simulations). Although both AI tools offer significant benefits, they also exhibit certain limitations and constraints: Data Coverage: The accuracy of the models depends heavily on the availability and quality of training data. Very novel or exotic chemical systems may require additional fine-tuning or supplemental datasets. Complex Systems: While MatterSim is capable of handling a wide range of conditions, highly complex or multi-phase systems may exceed the current accuracy limits of neural network-based simulations. Generalizability: The models are particularly effective for inorganic solid-state materials. However, support for organic compounds or hybrid interfaces remains limited at this stage. Hardware Requirements: Optimal performance requires modern GPUs and compatible software libraries (e.g., CUDA, PyTorch). Deployment on certain hardware architectures, such as Apple Silicon, is still considered experimental. Planned developments and future directions for Microsoft’s AI tools in materials science focus on expanding their capabilities and accessibility. Ongoing research and development aim to broaden the chemical space supported by the models, including more complex systems such as organic–metal interfaces. Integration with experimental robotics and automated feedback loops is expected to enhance closed-loop “design-build-test” workflows, accelerating discovery cycles. Additionally, educational modules are being developed to make these tools more accessible for teaching purposes, supported by tailored curricula and datasets. Finally, MatterGen and MatterSim are poised to enable rapid hypothesis testing in emerging fields such as quantum materials, advanced semiconductors, and energy storage technologies. 3 Methodology In this study, the aim is to utilize open-source platforms for capstone projects in materials development, both in educational and research contexts [43, 44]. The long-term goal is to integrate MatterGen and MatterSim into materials science workflows to significantly accelerate the discovery and validation of new materials. For effective integration, the following tasks are necessary [38, 39]: 1. Environment Setup and Installation Installation followed the official GitHub documentation and arXiv paper. MatterGen and MatterSim are available as open-source tools, implemented in Python and relying on PyTorch. Python 3.G Python 3.x (tested with version 3.9) was used. GPU support is strongly recommended for efficient computation. 2. Data Preparation Datasets containing material structures and property labels should be curated. While MatterGen can operate without labels, performance improves with labeled data for conditional generation. 3. Generative Design with MatterGen New material candidates can be generated using the pre-trained model. For targeted design, MatterGen may be fine-tuned on domain-specific datasets to guide generation toward desired compositions, symmetries, or property constraints, using approaches similar to ControlNet [45]. 4. High-Throughput Simulation with MatterSim MatterSim is used to illustrate an educational validation workflow on Fe and Al as canonical test systems. The tool validates physical stability and predicts properties of candidates under a wide range of conditions. 5. Iterative Workflow: Flywheel Integration MatterGen and MatterSim can be integrated into an iterative loop: thousands of structural candidates are generated with MatterGen, simulated and triaged using MatterSim, and refined based on experimental or computational feedback. This closed-loop workflow accelerates the identification of promising materials and reduces the scope of experimental validation. 6. Practical Applications This workflow is applicable to a wide range of use cases requiring novel materials. Both academic and industrial research teams have adopted this approach for rapid prototyping and project-driven materials selection. 7. Documentation, Community, and Support Comprehensive usage guides and Application Programming Interface (API) documentation are available via the respective GitHub repositories and official documentation portals. Active participation in community discussions and Q&A forums is encouraged to gain support and insights from peer use cases. The short-term objective of this study is to establish a computational environment in which the tools can be installed, tested, and effectively applied, along with a functional workflow for simulation tasks. The following tasks have been completed to support this goal: 1. Installation and setup of MatterSim and MatterGen In addition to following official installation procedures, our experience revealed practical challenges specific to university teaching environments: Hardware Considerations: Student laptops and university workstations often have limited GPU memory (4-8 GB). This constrains supercell sizes for phonon calculations and trajectory lengths for molecular dynamics simulations. We recommend minimum 8 GB GPU memory for educational use, with 16 GB preferred for advanced projects. Software Environment Issues: On Windows systems, Conda/Visual C++ Build Tools compatibility is critical. Students encountered issues with CUDA vs MPS (Apple Silicon) platforms, with numerical discrepancies in some calculations. Switching to CPU execution or CUDA-enabled systems resolved these issues. Realistic Simulation Scope: For semester-long projects, structural relaxation and phonon analysis are highly reliable. Molecular dynamics simulations are feasible for small systems. However, automated phase diagram generation requires additional validation and is best suited for advanced projects with instructor support. Timeline Expectations: Installation and environment setup typically require 2-4 hours. Initial validation simulations (Fe, Al) take 1-2 weeks. Full capstone projects span 8-12 weeks. To enable the use of MatterSim and MatterGen, a clean and isolated Python environment should be created. It is recommended to use Conda or Mamba to avoid version conflicts and to ensure proper integration of Visual C++ tools on Windows systems. The following procedures are necessary for implementation: a) Software requirements - Windows 10/11 with administrator rights - Python 3.G (MatterSim recommends version 3.6 for maximum stability) - Conda or Mamba installed (instructions: https://docs.conda.io/) - Visual C++ Build Tools (for Windows compilers) b) Create virtual environment c) Update Pip and build helper packages d) Install MatterSim - Stable version from PyPI - Developer version directly from GitHub - Installation from cloned source code e) Install MatterGen f) Final test 2. Code integration and workflows A Python-based graphical user interface (GUI) application was developed to facilitate atomistic simulations. The application is based on the MatterSim Machine Learning Force Field (MLFF) model, version v1.0.0-1M, and integrates several Python libraries, including the Atomic Simulation Environment (ASE), Phonopy for phonon calculations, and Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) for molecular dynamics simulations. The GUI is not simply a thin wrapper around existing libraries, but a didactic layer that encodes good-practice workflows for atomistic simulations in educational settings. The architecture consists of: Front-end: Built using standard Python widget frameworks (Tkinter/PyQt) with integrated plotting libraries (Matplotlib) for real-time visualization. The layout uses progressive disclosure, showing basic parameters by default with expandable panels for advanced options. Back-end: Orchestrates ASE calculator instantiation, MatterSim force-field loading, LAMMPS interface configuration, and data pipelines between components. The GUI manages supercell generation, displacement amplitude calculations for phonon analysis, ensemble choices (NVT/NPT) for molecular dynamics, and result formatting. Workflow Encoding: The GUI embeds validated workflows tested in our educational context: sensible defaults for displacement amplitudes (0.01 Å), supercell sizes (2×2×2 for metals), temperature ranges (0-1500 K), and timesteps (1 fs for MD). Human-Centric Design Principles are: (a) Sensible defaults to minimize decision paralysis for novice users (b) Inline documentation for each parameter with tooltips explaining physical meaning (c) Progressive disclosure of advanced options to avoid overwhelming beginners (d) Workflow templates for common tasks (structural relaxation, phonon calculation). The GUI is open-source and extensible, enabling instructors and students to adapt it to new materials and future versions of MatterSim and MatterGen. New figures will be added showing: (1) annotated screenshot with key GUI modules and workflows, (2) flowchart illustrating how the GUI orchestrates MatterSim, ASE, Phonopy, and LAMMPS. The GUI enables users to select crystal structures, perform computational tasks, and analyze simulation results. The application provides four main functional modules: - Structural Relaxation: Optimization of atomic positions within crystal structures to minimize total energy and achieve equilibrium geometry. - Phonon Calculations: Determination of phonon dispersion relations and the phonon density of states, which are essential for understanding thermal and vibrational properties. - Molecular Dynamics Simulations: Execution of simulations in canonical (NVT: constant number of particles, volume, and temperature) and isothermal-isobaric (NPT: constant number of particles, pressure, and temperature) ensembles, including temperature ramping and pressure variation studies. - Thermodynamic Property Estimation: Calculation of macroscopic quantities such as Gibbs free energy, entropy, and heat capacity based on atomistic simulation data. The first phase of the project focused on using MatterSim to simulate material properties, such as those of Fe, Al, and their alloys. The core functionalities, e.g. structural relaxation, force and stress calculations, and phonon analysis, were successfully executed during initial testing. In contrast, more advanced thermodynamic calculations were only partially supported in the open-source version of MatterSim utilized in this study. All simulations were carried out as planned, and the resulting data were systematically compared with values reported in the scientific literature. 4 Results and Discussion The most important findings of this study can be summarized as follows: The developed Python-based Graphical User Interface (GUI) enables intuitive execution of simulations with MatterSim and enhances the transparency of the underlying workflows for users. The essential computational tasks – structural relaxation, force and stress calculations, and phonon analysis – were successfully implemented. The results qualitatively align with expectations reported in the literature. Molecular dynamics simulations were feasible in selected cases; however, limitations were observed in the automated phase analysis functionality. Thermodynamic properties could only be estimated to a limited extent due to constraints in the open-source version of MatterSim used. The detailed findings will be presented and discussed below. 4.1 Python-based Graphical User Interface (GUI) A Python-based Graphical User Interface (GUI) was successfully developed to enable convenient use and execution of simulations. Parameters such as chemical element, lattice structure, lattice constant, temperature, and pressure can be entered. Simulation modes, including bulk relaxation, bulk phonon, molecule relaxation, and molecule phonon, can be executed. Various material properties, such as total energy, atomic forces and stress, heat capacity, and vapor pressure as a function of temperature, can be calculated. The developed GUI for visualization and simulation is demonstrated using examples of Fe (body-centered cubic structure, bcc) and Al (face-centered cubic structure, fcc), as shown in Fig. 1 . 4.2 Relaxation and force/stress calculation The relaxation function enables the optimization of crystal structures using MatterSim [ 39 , 44 , 46 ]. In this process, the MatterSim calculator is linked to an atomic structure object provided by the Atomic Simulation Environment (ASE). The structure is then relaxed using an optimization algorithm, such as the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method. As a result, the parameters – total energy, atomic forces, and stress tensors – were calculated and are presented in Fig. 1 for Fe (bcc structure) and Al (fcc structure), for example, at a pressure of 1 GPa and a room temperature of 300 K: The total energy E (in eV) represents the lattice potential energy modeled by MatterSim, which is the sum of all interatomic interactions. It reflects the energetic state of the atomic configuration and is fundamental for understanding the material’s stability and chemical properties [ 47 ]. In general, a lower total energy indicates a more stable atomic configuration, such as a crystal unit cell, a molecule, or an atomic cluster, where an energy minimum corresponds to a relaxed and stable structure. This work reports total energies of − 8.46 eV for idealized Fe (bcc structure) and − 3.72 eV for Al (fcc structure). Since the bcc Fe unit cell contains two atoms, the total energy of − 8.46 eV corresponds to approximately − 4.23 eV per atom. For fcc Al, the reported total energy already refers to a single atom. Therefore, the total energy normalized per atom for Fe (bcc) and Al (fcc) in this study is − 4.23 eV and − 3.72 eV, respectively. These values are in good agreement with the cohesive energies reported in the literature [ 46 ], which are defined as the energy required to separate neutral atoms in their ground electronic state from the solid at 0 K and 1 atm, and amount to − 4.28 eV for Fe and − 3.39 eV for Al. Atomic forces F (in eV/Å) are obtained by computing the gradient of the energy with respect to atomic positions via automatic differentiation [ 48 ]. These forces are essential for structural optimization and molecular dynamics simulations. The force vector (F = [Fx, Fy, Fz]) with a 1×3 output represents the net force acting on the considered atoms. Values close to 0 eV/Å indicate that the system is in force equilibrium. For relaxed metallic structures, residual forces typically range between 10⁻³ and 10⁻⁵ eV/Å [ 46 ]. The computed atomic forces for Fe and Al in this study are 0 eV/Å, confirming that the system is in force equilibrium. The stress tensor σ (in eV/Å or GPa) characterizes mechanical tensions within the material and is relevant for determining mechanical properties. The 3×3 matrix describes the stress tensor σ [ 46 , 49 ]: $$\:\sigma\:=\left[\begin{array}{ccc}{\sigma\:}_{xx}&\:{\sigma\:}_{xy}&\:{\sigma\:}_{xz}\\\:{\sigma\:}_{yx}&\:{\sigma\:}_{yy}&\:{\sigma\:}_{yz}\\\:{\sigma\:}_{zx}&\:{\sigma\:}_{zy}&\:{\sigma\:}_{zz}\end{array}\right]$$ 1 Diagonal elements (σxx, σyy, σzz) represent normal stresses, while off-diagonal elements (σxy, σxz, σyz …) represent shear stresses. In an ideal relaxed metal crystal without external loads, all components should be close to 0 GPa. This was confirmed by the results of this study, which illustrate both the structural stability and the predictive accuracy of the MatterSim model. 4.3 Phonon calculations Phonon calculations were performed using MatterSim in combination with Phonopy via the frozen-phonon method [ 48 , 50 ]. Phonopy generates supercells with small atomic displacements (typically ~ 0.01 Å), and MatterSim computes the corresponding atomic forces using its M3GNet-based machine-learning potential. These forces are then processed by Phonopy to construct the force-constant matrix, from which phonon dispersion relations, phonon density of states (DOS), and dynamical stability are derived. Rigorous benchmarking of MatterSim across thousands of materials has been reported in the official technical documentation and MatBench Discovery platform. Our results on Fe and Al demonstrate how this validated model can be translated into a student-friendly workflow. This approach enables efficient and accurate prediction of vibrational properties critical for thermal conductivity, heat capacity, and material stability. This method demonstrated strong performance in the tests, with the phonon DOS and dispersion spectra for Fe and Al presented in Fig. 2 – 3 . The phonon dispersions and densities generated in the GUI exhibit consistent behavior, confirming the physical validity of the results: no imaginary frequencies occur, indicating stable lattice structures; the acoustic and optical branches show the expected shapes; and the spectra qualitatively agree with known DFT reference data [ 51 , 52 ]. Comparison with published DFT results shows good agreement in phonon frequencies for Fe and Al and supports the use of MatterSim for thermal-property predictions in educational workflows. Table 2 compares key phonon properties between MatterSim predictions and published DFT references. Table 2 Quantitative comparison of phonon properties: MatterSim versus DFT Material Properties MatterSim DFT Reference Fe (bcc) Max frequency (THz) 8.2 8.4 [ 51 ] Debye temperature (K) 455 470 [ 51 ] Acoustic slope (km/s) 6.1 6.3 [ 51 ] Al (fcc) Max frequency (THz) 9.8 9.5 [ 52 ] Debye temperature (K) 425 428 [ 52 ] Acoustic slope (km/s) 6.4 6.5 [ 52 ] This high accuracy stems from MatterSim’s training on numerous distorted supercells, which enhances its ability to predict forces for phononic displacements. Consequently, phonon calculations rank among the most reliable features of the open-source MatterSim version. 4.4 Molecular dynamics (MD) In the Molecular Dynamics (MD) section of the Graphical User Interface (GUI), the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) acts as the integration engine, while MatterSim provides the interatomic forces [ 39 , 46 ]. This setup enables simulations in the canonical ensemble (NVT) and the isothermal-isobaric ensemble (NPT), such as temperature ramping or pressure variation runs. In practice, expected physical trends were observed; for instance, the internal energy increased with rising temperature. However, generating a fully automated phase diagram was not possible because the version of MatterSim used does not support this functionality. Nevertheless, the raw data from the graphical user interface can be utilized to plot temperature and energy curves that qualitatively indicate phase transitions. 4.5 Thermodynamic properties The Graphical User Interface (GUI) includes experimental features for evaluating thermodynamic properties such as free energy and heat capacity (Fig. 4 – 5 ). These calculations are based either on the phonon density of states or on results from molecular dynamics simulations, and are used to estimate quantities such as heat capacity at constant pressure ( \(\:{C}_{p}\) ) and enthalpy [ 39 , 44 ]. The heat capacity curves \(\:{C}_{p}\left(T\right)\:\) generated by the GUI exhibit trends that are partially plausible but also include some inconsistencies. Typically, the specific heat capacity \(\:{C}_{p}\:\) shows a sharp increase (or discontinuity) at the melting point due to the solid-to-liquid phase transition, which is confirmed by calculations for Fe and Al in this study. However, the \(\:{C}_{p}\:\) curves are only partially reliable and should be reported as a known limitation. These deviations arise from numerical issues in the GUI (due to differentiation of unsmoothed free-energy curves), limitations of MatterSim (missing electronic contributions and incomplete free-energy data), and noise introduced by small supercells in phonon DOS calculations. The calculated vapor pressure curves show the expected trend of increasing with higher temperatures. However, the absolute values should be treated with caution, which is not surprising given the limitations of the model. Determining vapor pressure requires accurate enthalpies of vaporization, realistic liquid and gas phases, and precise free energies [ 53 ]. MatterSim cannot provide these because it was not trained for gas phases, does not reliably represent liquids, and the open version only delivers approximate static energies. Additionally, the GUI approach, such as differentiation or Molecular Dynamics-based volume changes, does not allow a valid determination of vapor pressure. Therefore, vapor pressure cannot be accurately computed using the open-source MatterSim version. Moreover, due to limitations in the open-source version of MatterSim used, comprehensive thermodynamic analyses were not fully supported. Consequently, only partial results could be obtained. For example, the calculated phase diagram of Fe (Fig. 6 ) indicated tendencies of possible phase transitions, such as from the fcc structure to the bcc structure. Nevertheless, a complete determination of the phase diagram was not feasible. The main reasons include incomplete Gibbs free energy data (electronic, configurational, and high-temperature contributions are missing), lack of Application Programming Interface (API) support (since the open-source MatterSim version does not provide integrated phase diagram functionality), and a highly simplified data basis with only a few temperature and pressure points (resulting in irregular and unstable diagrams). Despite these limitations, the calculated phase diagram of Fe is partly consistent with reference data (Fig. 7 ) [ 54 ]. Furthermore, attempts to create phase diagrams for Fe–C and Al–Si alloys were unsuccessful. Only eutectic points (Al–Si at 577°C and Fe–C at 1147°C) as well as the eutectoid point of Fe–C at 723°C could be confirmed in this study. Nevertheless, complete phase diagrams for Fe–C and Al–Si alloys could not be generated; therefore, the phase diagrams of these alloys are not reported here. 4.5 Limitations and Challenges The practical implementation of the simulations revealed three key limitations that should be considered in future work involving MatterSim: 1. Computational and Storage Resources: Large supercells, extended molecular dynamics simulations, and especially the training or fine-tuning of large-scale models demand substantial Graphics Processing Unit (GPU) memory. In the absence of sufficient hardware resources, limitations in batch size, supercell dimensions, and model complexity can reduce the significance and reliability of certain results. 2. Infrastructure and Platform Dependency: Managing large model weights and trajectory datasets requires robust data transfer and version control mechanisms. Unstable network connections or the absence of a Large File Storage (LFS) infrastructure can delay workflows. Additionally, numerical discrepancies may arise between computing platforms (e.g., Metal Performance Shaders (MPS) on Apple Silicon versus Compute Unified Device Architecture (CUDA) on x86 systems). In problematic cases, switching to Central Processing Unit (CPU) execution or alternative hardware platforms may be necessary. 3. Methodological Limitations and Validation: The publicly available version of MatterSim used in this project supports reliable structural relaxations, force calculations, and phonon analyses. However, it does not provide full support for automated and comprehensive thermodynamic workflows, such as the generation of complete phase diagrams. Machine learning-based force fields are inherently data-driven and may exhibit systematic deviations when applied outside their training domain. Therefore, targeted validation using Density Functional Theory (DFT) or experimental data is essential. In this project, selected predictions were cross-validated using DFT, with deviations typically within the single-digit percentage range. MatterGen was not used productively in this phase of the project. For generative applications, targeted training or fine-tuning on appropriate datasets and the development of a validation pipeline are still required. 5 Conclusions and Outlook This study explores the integration of the open-source platforms MatterGen and MatterSim into a computational workflow for materials development in educational and research settings. A Python-based Graphical User Interface (GUI) was developed to support key simulation tasks, including structural relaxation, phonon analysis, molecular dynamics, and thermodynamic property estimation. Phonon spectra generated using MatterSim in combination with Phonopy showed good agreement with Density Functional Theory (DFT) reference data. Molecular dynamics simulations, powered by the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS), revealed expected physical trends, although automated phase diagram generation was not supported. Thermodynamic properties such as free energy and heat capacity were partially estimated, with qualitative trends successfully reproduced. Challenges encountered included limitations in hardware resources (especially Graphics Processing Unit memory), infrastructure dependencies (such as the need for Large File Storage systems), and methodological constraints in the open-source version of MatterSim. While MatterGen was not actively used in this phase, it shows strong potential for future generative design tasks. Short-term recommendations include leveraging GPU cluster resources, ensuring stable data infrastructure (e.g., Large File Storage), integrating free energy methods such as the Quasi-Harmonic Approximation and thermodynamic integration, and validating results systematically using Density Functional Theory. Long-term potential lies in combining generative models with machine learning-based force fields to enable efficient materials screening workflows. Future directions involve exploring MatterGen and merging generative and predictive models for rapid screening. This work highlights AI’s role as a valuable resource for both teaching materials engineering and accelerating computational materials discovery. The custom Python-based GUI, integrated with MatterSim, is available as open source on GitHub for public access and further development ( https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git ). Abbreviations AI Artificial Intelligence ASE Atomic Simulation Environment API Application Programming Interface BFGS Broyden–Fletcher–Goldfarb–Shanno CNN Convolutional Neural Network DFT Density Functional Theory DOS Density of States FEM Finite Element Method FCC Face-Centered Cubic BCC Body-Centered Cubic GPU Graphics Processing Unit GUI Graphical User Interface LAMMPS Large-scale Atomic/Molecular Massively Parallel Simulator LFS Large File Storage ML Machine Learning MLFF Machine Learning Force Field MPS Metal Performance Shaders NPT Isothermal-Isobaric Ensemble (constant Number of particles, Pressure, Temperature) NVT Canonical Ensemble (constant Number of particles, Volume, Temperature) PV Photovoltaic R&D Research and Development SUN Stable, Unique, and Novel Declarations Acknowledgments This manuscript is an extended version of the paper ‘Artificial Intelligence in Materials Science Education and Research & Development’ , originally presented at The 2025 International Scientific Conference on Media Education in the age of AI and Digital Transformation on 25 December 2025 in Hanoi, Vietnam. The authors would like to thank the organizers of the conference for providing the opportunity to present the preliminary version of this work and for granting permission to reuse the content of the published conference paper for submission to another journal. Author Contributions The authors confirm their contributions to the paper as follows: Conceptualization and methodology: Gia Khanh Pham, Ulrick Edwin Nguegang Tsopmo, Marta Segura Cubells. Software: Ulrick Edwin Nguegang Tsopmo. Data collection and analysis: Ulrick Edwin Nguegang Tsopmo, Gia Khanh Pham. Draft manuscript: Gia Khanh Pham, Ulrick Edwin Nguegang Tsopmo. Manuscript revisions: Ulrick Edwin Nguegang Tsopmo, Marta Segura Cubells. Supervision: Gia Khanh Pham. All authors have read and approved the final version of the manuscript for publication. Funding No funding was received for conducting this study. Data Availability The data from this study and the open-source version of the GUI are available on GitHub (https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git) or can be obtained from the corresponding author upon reasonable request. Conflict of interest The authors declare no conflicts of interest. Ethics Approval This manuscript is an extended version of a conference paper. The authors confirm that this work complies with the ethical responsibilities and standards for research and publication. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. References Nematov, D., & Hojamberdiev, M. (2025). Machine learning–driven materials discovery: Unlocking next-generation functional materials – A review. Computational Condensed Matter, 45 (2025), 1-23. https://doi.org/10.1016/j.cocom.2025.e01139 Yılmaz, Ö. (2024). Personalised learning and artificial intelligence in science education: Current state and future perspectives. Educational Technology Quarterly, 2024(3), 255–274. https://doi.org/10.55056/etq.744 Guinan, G., Salvador, A., Smeaton, M. A., Glaws, A., Egan, H., Wyatt, B. C., Anasori, B., Fiedler, K. R., Olszta, M. J.,& Spurgeon, S. R. (2025). Mind the gap: Bridging the divide between AI aspirations and the reality of autonomous characterization. APL Mach. Learn, 3(2), 020903. https://doi.org/10.1063/5.0267699 Tom, G., Schmid, S. P., Baird, S. G., Cao, Y., Darvish, K., Hao, H., Lo, S., Pablo-García, S., Rajaonson, E. M., Skreta, M., Yoshikawa, N., Corapi, S., Akkoc, G. D., Strieth-Kalthoff F., Seifrid, M., & Aspuru-Guzik, A. (2024). Self-driving laboratories for chemistry and materials science. Chemical Reviews, 124, 9633−9732. https://doi.org/10.1021/acs.chemrev.4c00055 Ma, J., Cao, B., Dong, S., Tian Y., Wang M., Xiong J., & Sun S. (2024). MLMD: A programming-free AI platform to predict and design novel materials. npj Computational Materials, 10 (59). https://doi.org/10.1038/s41524-024-01243-4 Gu, Z., Tran, H., Manjunath, A., Chen, D., & Fu, S. (2024). AI automated microstructure analysis for intelligent manufacturing. Open Journal of Robotics and Automation Technology, 3(2), 110–122. https://irispublishers.com/ojrat/pdf/OJRAT.MS.ID.000533.pdf Na, J., Kim, S. J., Kim, H., Kang, S. H., & Lee, S. (2023). A unified microstructure segmentation approach via human-in-the-loop machine learning. Acta Materialia, 255. https://doi.org/10.1016/j.actamat.2023.119086 Pham, G. K., Yalcin, K., Binti Wan, A. W. A. N., Nguyen, A. S., & Trinh, V. T. (2024). Development of an Open-Source Software Tool for Microstructure Analysis of Materials Using Artificial Intelligence. Key Engineering Materials, 1004, 103–110. https://doi.org/10.4028/p-vcog1h Kabudi, T., Pappas, I., & Olsen, D. H. (2021). AI-enabled adaptive learning systems: A systematic mapping of the literature. Computers and Education: Artificial Intelligence, 2. https://doi.org/10.1016/j.caeai.2021.100017 Katona, J., & Gyonyoru, K. I. K. (2025). Integrating AI-based adaptive learning into the flipped classroom model to enhance engagement and learning outcomes. Computers and Education: Artificial Intelligence, 8. https://doi.org/10.1016/j.caeai.2025.100392 eLearning Industry. (2024). Adaptive learning with AI: Revolutionizing personalized education. eLearning Industry. https://elearningindustry.com/understanding-adaptive-learning-how-ai-is-revolutionizing-personalized-education (accessed on 20.10.2025) Liu, Y., Yang, Z., Yu, Z., Liu, Z., Liu, D., Lin, H., Li, M., Ma, S., Avdeev, M., Shi, S. (2023). Generative artificial intelligence and its applications in materials science: Current situation and future perspectives. Journal of Materiomics, 9(4), 798-816. https://doi.org/10.1016/j.jmat.2023.05.001 Kumar, K. (2025). Generative AI in materials science: Accelerating discovery through inverse design. SSRN Paper No. 5251184. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5251184 Erduran, S., & Levrini, O, (2024). The impact of artificial intelligence on scientific practices: an emergent area of research for science education. International Journal of Science Education, 46(18), 1982–1989. https://doi.org/10.1080/09500693.2024.2306604 DeCost, B. L., Hattrick-Simpers, J. R., Trautt, Z., Kusne, A. G., Campo, E., & Green, M. L. (2020). Scientific AI in materials science: A path to a sustainable and scalable paradigm. Mach. Learn.: Sci. Technol. 1, 033001. https://doi.org/10.1088/2632-2153/ab9a20 Batool, A., Sanumi, O., & Jankovic, J. (2024). Application of artificial intelligence in the materials science, with a special focus on fuel cells and electrolyzers. Energy and AI, 18. https://doi.org/10.1016/j.egyai.2024.100424 Peivaste, I., Belouettar, S., Mercuri, F., Fantuzzi, N., Dehghani, H., Izadi, R., Ibrahim, H., Lengiewicz, J., Belouettar-Mathis, M., Bendine, K., Makradi, A., Horsch, M., Klein, P., El Hachemi, M., Preisig, H. A., Rezgui, Y., Konchakova, N., & Daouadji, A. (2025). Artificial intelligence in materials science and engineering: Current landscape, key challenges, and future trajectories. Composite Structures, 372, 119419. https://doi.org/10.1016/j.compstruct.2025.119419 Bai, X., & Zhang, X. (2025). Artificial Intelligence-Powered Materials Science. Nano-Micro Lett. 17, 135. https://doi.org/10.1007/s40820-024-01634-8 Chitturi, S. R., Ramdas, A., Wu, Y., Rohr, B., Ermon, S., Dionne, J., da Jornada, F. H., Dunne, M., Tassone, C., Neiswanger, W., & Ratner, D. (2024). Targeted materials discovery using Bayesian algorithm execution. npj Computational Materials, 10, 156. https://doi.org/10.1038/s41524-024-01326-2 Merchant, A., Batzner, S., Schoenholz, S.S., Aykol, M., Cheon, G., Cubuk, E. D. (2023). Scaling deep learning for materials discovery. Nature, 624, 80–85. https://doi.org/10.1038/s41586-023-06735-9 Priyadarshini, M. S., Romiluyi, O., Wang, Y., Miskin, K., Ganley, C., & Clancy, P. (2024). PAL 2.0: A physics driven Bayesian optimization framework for material discovery. Materials Horizons, 11(28), 781–791. https://doi.org/10.1039/d3mh01474f Alrfou, K., Kordijazi, A., & Zhao, T. (2022). Computer vision methods for the microstructural analysis of materials: The State-of-the-art and Future Perspectives. arXiv:2208.04149. https://doi.org/10.48550/arXiv.2208.04149 Mishra, S. P., & Rahul, M. R. (2023). A detailed study of convolutional neural networks for the identification of microstructure. Materials Chemistry and Physics, 308, 128275. https://doi.org/10.1016/j.matchemphys.2023.128275 Gao, S., Huang, Q., Huang, C., Li, C., Liu, K., Sa, B., Yu, Y., Xue, D., Liu, Z., & Dai, M. (2025). Deep generative model for the inverse design of Van der Waals heterostructures. Sci Rep 15, 23023. https://doi.org/10.1038/s41598-025-06432-9 Qin, C., Liu, J., Ma, S., Du, J., Jiang, G., & Zhao, L. (2024). Inverse design of semiconductor materials with deep generative models. Journals of Materials Chemistry A, 12, 22689-22702. https://doi.org/10.1039/d4ta02872d Papadimitriou, I., Gialampoukidis, I., Vrochidis, S., & Kompatsiaris, I. (2024). AI methods in materials design, discovery and manufacturing: A review. Computational Materials Science, 235, 112793. https://doi.org/10.1016/j.commatsci.2024.112793 Xu, P., Ma, Y., Lu, W., Li, M., Zhao, W., & Dai, Z. (2025). Multi-objective optimization in machine learning assisted materials design and discovery. J. Mater. Inf. 2025, 5, 26. http://dx.doi.org/10.20517/jmi.2024.108 Jiang, X., Xue, D., Bai, Y., Wang, W. Y., Liu, J., Yang, M., & Su, Y. (2025). AI4Materials: Transforming the landscape of materials science and engineering. Review of Materials Research, 1(1), 100010. https://doi.org/10.1016/j.revmat.2025.100010 Merchant, A., & Cubuk, E. D. (2023). Millions of new materials discovered with deep learning. Google DeepMind. https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/ (accessed on 20.10.2025) Szymanski, N. J., Rendy, B., Fei, Y., Kumar, R. E., He, T., Milsted, D., McDermott, M. J., Gallant, M., Cubuk, E. D., Merchant, A., Kim, H., Jain, A., Bartel, C. J., Persson, K., Zeng, Y., & Ceder, G. (2023). An autonomous laboratory for the accelerated synthesis of novel materials. Nature 624, 86–91. https://doi.org/10.1038/s41586-023-06734-w Mianroodi, J.R., H. Siboni, N. & Raabe, D. (2021). Teaching solid mechanics to artificial intelligence – a fast solver for heterogeneous materials. npj Comput Mater 7, 99. https://doi.org/10.1038/s41524-021-00571-z Trewartha, A., & Torrisi, S. (2024). Toyota Research Institute to Launch Challenge to Accelerate Research in New Advanced Materials. Toyota Research Institute. https://medium.com/toyotaresearch/toyota-research-institute-to-launch-program-to-accelerate-research-in-new-advanced-materials-d52d5d65e8d9 (accessed on 24.10.2025) Mulholland, G. J. (2018). AI for accelerated materials discovery. Citrine. https://worldmaterialsforum.com/files/Presentations2018/PS5/WMF%202018%20Citrine%20Informatics%20Mulholland.pdf (accessed on 24.10.2025) Fraunhofer IWM (2025). KI-Konzepte für Materialentwicklung und Werkstofftechnologie. https://www.iwm.fraunhofer.de/de/geschaeftsfelder/bauteilsicherheit-und-leichtbau/meso-und-mikromechanik/KI-konzepte_materialentwicklung_werkstofftechnologie.html (accessed on 20.10.2025) DFKI Deutsches Forschungszentrum für Künstliche Intelligenz (2022). Machine Learning für die Herstellungsprozess-Simulation für Verbundwerkstoffe. https://www.dfki.de/web/news/machine-learning-simulation-verbundwerkstoffe (accessed on 20.10.2025) Wu, J., Torresi, L., Hu, M., Reiser, P., Zhang, J., Rocha-Ortiz, J. S., Wang, L., Xie, Z., Zhang, K., Park, B., Barabash, A., Zhao, Y., Luo, J., Wang, Y., Lüer, L., Deng, L.-L., Hauch, J. A., Guldi, D. M., Pérez-Ojeda, M. E., Seok, S. I., Friederich, P., & Brabec, C. J. (2024). Inverse design workflow discovers hole-transport materials tailored for perovskite solar cells. Science, 386(6727), 1256-1264(2024). https://doi.org/10.1126/science.ads0901 Microsoft (2025). AI meets materials discovery: The vision behind MatterGen and MatterSim. https://www.microsoft.com/en-us/research/story/ai-meets-materials-discovery/ (accessed on 24.10.2025) Zeni, C., Pinsler, R., Zügner, D., Fowler, A., Horton, M., Fu, X., Wang, Z., Shysheya, A., Crabbé, J., Ueda, S., Sordillo, R., Sun, L., Smith, J., Nguyen, B., Schulz, H., Lewis, S., Huang, C.-W., Lu, Z., Zhou, Y., Yang, H., Hao, H., Li, J., Yang, C., Li, W., Tomioka, R., & Xie, T. (2025). A generative model for inorganic materials design. Nature 639, 624–632. https://doi.org/10.1038/s41586-025-08628-5 Yang, H., Hu, C., Zhou, Y., Liu, X., Shi, Y., Li, J., Li, G., Chen, Z., Chen, S., Zeni, C., Horton, M., Pinsler, R., Fowler, A., Zügner, D., Xie, T., Smith, J., Sun, L., Wang, Q., Kong, L., Liu, C., Hao, H., & Lu, Z. (2024). MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures. arXiv:2405.04967. https://doi.org/10.48550/arXiv.2405.04967 Pham, G. K., Nguegang Tsopmo, U. E., Segura Cubells, M. (2025). Artificial Intelligence in Materials Science Education and Research & Development. The 2025 International Scientific Conference on Media Education in the age of AI and Digital Transformation, 25.12.2025, Hanoi, Vietnam. Conference proceedings, ISBN: 978-604-45-2379-8. Swayne M. (2025). What is MatterGen? Microsoft’s generative AI could transform materials research. https://theaiinsider.tech/2025/01/17/what-is-mattergen-microsofts-generative-ai-could-transform-materials-research/ (accessed on 24.10.2025) Xie, T. (2024). MatterGen: A Generative Model for Materials Design. https://www.microsoft.com/en-us/research/articles/mattergen-a-generative-model-for-materials-design/ (accessed on 24.10.2025) Microsoft/MatterGen. (2024). MatterGen [Software]. GitHub. https://github.com/microsoft/mattergen (accessed on 24.10.2025) Microsoft/MatterSim. (2024). MatterSim [Software]. GitHub. https://github.com/microsoft/mattersim (accessed on 24.10.2025) Zhang, L., Rao, A., Agrawala, M. (2023). Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 3836–3847. https://doi.org/10.1109/ICCV51070.2023.00355 Microsoft/MatterSim. (2024). MatterSim Documentation: Getting Started. GitHub. https://microsoft.github.io/mattersim/user_guide/getting_started.html (accessed on 14.11.2025) Kittel, C. (2005). Introduction to solid state physics (8th ed.). Hoboken, NJ: Wiley. Phonopy (2025). Workflow. https://phonopy.github.io/phonopy/workflow.html (accessed on 15.11.2025) Ibach, H., Lüth, H. (2009). Festkörperphysik: Einführung in die Grundlagen (7. Aufl.). Berlin: Springer. DeepWiki (2025). Microsoft/mattersim: Phonon Calculations. https://deepwiki.com/microsoft/mattersim/6.2-phonon-calculations (accessed on 15.11.2025) CASTEP Documentation (2023). Calculating phonon spectra for ferromagnetic iron. https://www.tcm.phy.cam.ac.uk/castep/documentation/WebHelp/content/tutorials/casteptutorial/castep_fdphonons_tut.htm (accessed on 15.11.2025) Sade, Y. A., Babaji, G., Lawal, M. A. (2017). Phonon dispersion relation and density of states in some selected FCC metal crystals. IOSR Journal of Applied Physics, 9(5), 47–51. https://doi.org/10.9790/4861-0905024751 Mondal, B., Mukherjee, T., Finch, N. W., Saha, A., Gao, M. Z., Palmer, T. A., DebRoy, T. (2023). Vapor pressure versus temperature relations of common elements. Materials, 16(50), 12345. https://doi.org/10.3390/ma16010050 Läpple, V., Drube, B., Wittke, G., Kammer, C. (2015). Werkstofftechnik Maschinenbau. Haan-Gruiten: Verlag Europa-Lehrmittel. ISBN 978-3-8085-5265-0. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 06 Apr, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviewers invited by journal 18 Mar, 2026 Submission checks completed at journal 10 Feb, 2026 First submitted to journal 09 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8143147","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":628562429,"identity":"116b5b1b-b8a1-4e3c-8ed3-68a1b2431a69","order_by":0,"name":"Gia Khanh Pham","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA8klEQVRIiWNgGAWjYBAC/mYGBsbGBgYGAwYGNiDfJgEkKoFPi8RhVC1phLUAVaJoOUyEFnbmB4wzd9jlmTPwHnvwM+d8nm4D88HbPPi0MLMZMG48k1xs2cCXbti77Xax2QG2ZGt8WgwPM5j/fNjGnLjhAI+ZBO+224nbgAxpvLYcZv/A+LCtHqxF8u+2c0At/N8IaOEBOqztMFiLNO+2AyBb2PBqkTjMU8A4s+144obDfGnSstuSE7cdZjO2nINHC3//8Q2MvW3ViRuO9x6TfLvNLnHb8eaHN97g0YIAzDDHMBOlHAzwuX8UjIJRMApGNAAAkM9QEdHukAwAAAAASUVORK5CYII=","orcid":"","institution":"Munich University of Applied Sciences","correspondingAuthor":true,"prefix":"","firstName":"Gia","middleName":"Khanh","lastName":"Pham","suffix":""},{"id":628562430,"identity":"3cbd2dd1-7dbd-4b47-8d66-74963d220eb1","order_by":1,"name":"Ulrick Edwin Nguegang Tsopmo","email":"","orcid":"","institution":"Munich University of Applied Sciences","correspondingAuthor":false,"prefix":"","firstName":"Ulrick","middleName":"Edwin Nguegang","lastName":"Tsopmo","suffix":""},{"id":628562431,"identity":"ce7593b7-b3c1-4993-a8d6-a73f0ccb104f","order_by":2,"name":"Marta Segura Cubells","email":"","orcid":"","institution":"Universidad Politécnica de Madrid","correspondingAuthor":false,"prefix":"","firstName":"Marta","middleName":"Segura","lastName":"Cubells","suffix":""}],"badges":[],"createdAt":"2025-11-18 08:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8143147/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8143147/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108383758,"identity":"d73db3df-005a-436c-b48f-930c7fb2952e","added_by":"auto","created_at":"2026-05-04 05:48:21","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":200252,"visible":true,"origin":"","legend":"\u003cp\u003eGraphical User Interface (GUI) for bulk relaxation calculations, showing outputs for total energy, atomic forces, and stress tensors: Fe (bcc structure) on the left and Al (fcc structure) on the right\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/1acebb9b81bcaa64eaab9810.png"},{"id":108383759,"identity":"613bdd8c-5d4b-4763-9198-a12ec9365c5e","added_by":"auto","created_at":"2026-05-04 05:48:21","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":48568,"visible":true,"origin":"","legend":"\u003cp\u003eCalculated phonon density of states (DOS) and phonon dispersion relation of Fe.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/aaa880ba15b9634480af98d7.png"},{"id":108383760,"identity":"916a8158-3ed4-4df8-966f-62165c6b9673","added_by":"auto","created_at":"2026-05-04 05:48:21","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":45466,"visible":true,"origin":"","legend":"\u003cp\u003eCalculated phonon density of states (DOS) and phonon dispersion relation of Al.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/ede13ed71479088f03b8f9ab.png"},{"id":108493302,"identity":"a9811c39-f082-41e9-b67f-04414c82f924","added_by":"auto","created_at":"2026-05-05 09:59:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":103969,"visible":true,"origin":"","legend":"\u003cp\u003eGUI for the calculation of material properties, such as vapor pressure and heat capacity, at various temperatures for Fe.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/7c626b741402bb46e929a177.png"},{"id":108803880,"identity":"1e80eb85-2002-4a8f-9bb2-25274e8bb508","added_by":"auto","created_at":"2026-05-08 15:09:55","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":122920,"visible":true,"origin":"","legend":"\u003cp\u003eGUI for calculating vapor pressure and heat capacity of Al at various temperatures.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/e8ada9073e667c0a9be91335.png"},{"id":108383763,"identity":"8729a38a-fc46-4e3f-828d-cecf610e41dd","added_by":"auto","created_at":"2026-05-04 05:48:21","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":167705,"visible":true,"origin":"","legend":"\u003cp\u003eCalculated phase diagram of Fe using MatterSim\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/872f714dd5a18a4005c4d3af.png"},{"id":108493588,"identity":"30a65fde-dd9b-4df1-b8df-b00fa01ed9ea","added_by":"auto","created_at":"2026-05-05 10:01:01","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":91988,"visible":true,"origin":"","legend":"\u003cp\u003eComparative phase diagram of Fe based on reference data from the literature (image adapted from the original) [54].\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/99a9e9bd703ce6a21e4db5ef.png"},{"id":108808980,"identity":"403b9973-8545-491a-ad38-6a79a02fd2b0","added_by":"auto","created_at":"2026-05-08 15:48:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1039896,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8143147/v1/50716fcb-0ab6-4984-a5b2-4c3e939b1f00.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Accelerating Computational Material Discovery and Learning: First Experiences with MatterSim","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eArtificial Intelligence (AI) plays a game-changing role in materials science by offering added value for education and research \u0026amp; development (R\u0026amp;D). Its integration into academic and industrial contexts enables significant advancements in data-driven methodologies and personalized learning environments [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Key applications of AI in materials science education and general pedagogy include:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eVirtual Labs \u0026amp; Simulations: AI enables realistic modeling of complex phenomena like microstructure evolution and fatigue, enhancing hands-on learning [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePredictive Materials Design: Machine learning supports property prediction and accelerates the development of novel materials [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAutomated Microstructure Analysis: AI tools classify phases and detect defects in material images, streamlining experimental evaluation [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eAdaptive Learning Systems: Intelligent platforms personalize content based on learners\u0026rsquo; progress and knowledge levels [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eGenerative AI for Teaching: AI assists in creating tailored exercises, assessments, and explanations for advanced topics [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eAI tools in materials science education offer benefits in personalized learning, enabling students to progress at their own pace with tailored feedback. They enhance data literacy by providing hands-on experience with real-world datasets and bridge the gap between theory and practice through simulations without costly lab setups [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. Students can solve problems faster by using AI for property prediction and process optimization, and they gain industry-relevant skills that prepare them for modern, data-driven R\u0026amp;D environments [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eBeyond education, Artificial Intelligence (AI) has become increasingly prominent in the R\u0026amp;D of novel materials. Numerous case studies and projects have demonstrated its applicability across several key domains [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Machine learning techniques facilitate the prediction of material properties, including mechanical, thermal, and electrical characteristics [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. In the filed of material discovery, Bayesian optimization and graph neural networks are employed to identify promising new compounds, while surrogate models significantly accelerate high‑throughput simulations [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Convolutional Neural Networks (CNNs) contribute to microstructure analysis [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e], and generative models enable the inverse design of materials tailored to specific target properties [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Process optimization benefits from Bayesian and semi‑supervised learning approaches [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], whereas the development of sustainable materials leverages multi‑objective optimization to balance performance metrics [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eApplying AI-driven methodologies in materials R\u0026amp;D has the potential to significantly shorten development cycles and foster the creation of novel material solutions [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents practical examples from industry and academia, demonstrating how artificial intelligence contributes measurable value and yields remarkable outcomes in materials R\u0026amp;D.\u003c/p\u003e \u003cp\u003e​\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eIndustry and Research Case Studies in AI-Driven Materials Development.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOrganization\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGoal\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI Approach\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eResult\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGoogle DeepMind\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMaterials discovery (batteries, semiconductors, catalysis)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGNoME (Graph Neural Networks for Materials Exploration)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2.2\u0026nbsp;million new crystals predicted; ~380,000 identified as stable; 736 already synthesized [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeta FAIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLarge-scale materials datasets and surrogate models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eOpen Materials 2024 (OMat24) dataset; Graph Neural Networks\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e118\u0026nbsp;million DFT calculations on 52\u0026nbsp;million structures; enables rapid property screening\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAcademic Autonomous Lab Lawrence Berkeley National Lab (LBNL) \u0026ndash; A-Lab\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAutonomous synthesis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSelf-driving lab: robotics\u0026thinsp;+\u0026thinsp;AI decision-making\u0026thinsp;+\u0026thinsp;ML-based characterization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eIn 17 days, 41 new compounds synthesized from 58 targets; lab operates 50\u0026ndash;100\u0026times; faster than manual workflows [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e].\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMax Planck Institute for Iron Research (MPIE) + DeepMetis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMechanics simulation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDeep-learning surrogate solver (DNN)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e8,300 times faster than standard FEM solvers for elasto-plastic materials [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eToyota Research Institute (TRI) \u0026amp; Partners (Stanford/MIT)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eBattery R\u0026amp;D\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eBayesian optimization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFaster lifetime prediction [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCitrine Informatics\u0026thinsp;+\u0026thinsp;HRL Laboratories (Boeing \u0026amp; GM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDesign of 3D-printable alloys for aerospace.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML screening\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ecycle-time reduction from years to days for candidate search \u0026amp; screening [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFraunhofer IWM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eData-scarce scenarios\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSemi-supervised learning\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLower annotation cost [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLeibniz-IVW\u0026thinsp;+\u0026thinsp;DFKI\u0026thinsp;+\u0026thinsp;Fraunhofer ITWM (ML4SIM)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFiber composite manufacturing process\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML-accelerated simulations (surrogates), anomaly detection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eFaster, robust process design [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eForschungszentrum J\u0026uuml;lich\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePerovskite PV \u0026ndash; inverse design of organic hole transport materials\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHigh-throughput synthesis\u0026thinsp;+\u0026thinsp;ML for rapid optimization\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNew molecules enhancing photovoltaic efficiency identified within weeks [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eA highly impressive example of AI tool development for R\u0026amp;D in new materials comes from Microsoft. Microsoft has developed two advanced AI tools, MatterSim and MatterGen, that are transforming materials science through rapid simulation and generative design [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMatterGen is a generative model for materials design [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. It can create novel inorganic material structures from scratch, tailored to user-defined property targets (such as hardness, conductivity, or magnetism), using a prompt-driven approach. MatterSim is a deep learning model for simulating and predicting material properties, such as mechanical, electronic, and thermodynamic behavior, under real-world conditions across a wide range of temperatures and pressures [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. It evaluates material candidates efficiently and with accuracy comparable to first-principles methods.\u003c/p\u003e \u003cp\u003eBoth MatterGen and MatterSim have been freely available since early 2024 as open-source projects and are accessible via Microsoft\u0026rsquo;s Azure Quantum Elements platform. These tools offer substantial benefits for both academic and industrial research [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. In academic settings, they help accelerate prototyping by reducing the number of costly experiments, thanks to their ability to efficiently narrow down candidate materials. In industrial R\u0026amp;D, they enable fast screening and prioritization of materials for applications in manufacturing, energy, and electronics.\u003c/p\u003e \u003cp\u003eFor instance, researchers can use these tools to design and simulate new electrode compositions for battery materials. In the development of magnets and fuel cells, they allow for the optimization of structures and the prediction of performance before synthesis. For photovoltaic applications, MatterGen generates candidate materials, and MatterSim assesses their thermal and mechanical stability over typical PV operating temperatures (e.g., -40\u0026deg;C to +\u0026thinsp;85\u0026deg;C). However, optical properties and light-matter interactions (e.g., bandgap engineering, absorption spectra) require complementary electronic-structure calculations (DFT, GW, Bethe-Salpeter equation) or device-level simulations. Additionally, they support rapid calculation of phase diagrams, enabling the prediction of material stability and transformations under varying conditions. The novelty of this work lies in (i) implementing and open-sourcing a human-centric GUI for MatterSim tailored to teaching and capstone projects, (ii) critically assessing which simulation tasks are reliable enough for educational use, and (iii) documenting practical limitations and validation strategies against DFT and reference data in a realistic university environment. Our contribution is positioned at the interface between cutting-edge AI models and their practical adoption in educational and applied research contexts, with explicit documentation of usability, limitations, and best-practice workflows.\u003c/p\u003e \u003cp\u003eThis study explores the application of advanced AI tools, MatterSim and MatterGen, for research and education in materials science. [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]. The focus of the project lies in the evaluation and analysis of these tools, aiming to assess their practical suitability for materials development.\u003c/p\u003e \u003cp\u003eWhile MatterSim and MatterGen are well documented and benchmarked by Microsoft, there remains a gap between these advanced tools and their practical integration into human-centric educational settings and project-based learning in materials science. Recent work on AI-enabled virtual labs and adaptive learning emphasizes the importance of such tools in education but also highlights an implementation gap in real teaching practice. Our aim is to: (i) bridge this gap by demonstrating how MatterSim can be integrated into a capstone project and course setting, (ii) provide an open-source GUI and workflow that lowers entry barriers for students without strong Python or command-line expertise, and (iii) document where current tools are robust enough for teaching and where their limitations require instructor guidance or complementary DFT validation.\u003c/p\u003e \u003cp\u003eThe primary objectives of the study include the installation and configuration of the tools, the execution of test simulations, and the assessment of simulation accuracy. In addition, aspects such as accessibility, usability, and computational efficiency are examined. An important task of the project is to create a graphical user interface (GUI) for convenient use and to document the entire workflow \u0026ndash; from installation to application \u0026ndash; in a way that enables other students and researchers to begin using these tools immediately, without significant setup time or technical barriers.\u003c/p\u003e \u003cp\u003eWhile both tools are considered, the first phase of the project places particular emphasis on an in-depth analysis and practical application of MatterSim, due to its capabilities for simulating material properties under realistic conditions. The study aims to contribute to the broader understanding of how AI-driven simulation can enhance both academic learning and industrial R\u0026amp;D processes in materials science.\u003c/p\u003e"},{"header":"2 Theoretical Foundations of Microsoft’s AI Tools: MatterSim and MatterGen","content":"\u003cp\u003eMatterGen uses a \u003cem\u003egenerative diffusion model\u003c/em\u003e which jointly predicts atomic structure, elements, and lattice parameters, fine-tunable for property constraints [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. MatterGen is like a \"creative AI scientist\" that designs entirely new materials from scratch, using advanced generative algorithms. Traditional methods can only select or filter from materials that are already known; MatterGen, however, can imagine materials that have never existed, following a multi-step process [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]:​\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eDiffusion-Based Generation: The process starts with a random cloud of atoms in space. MatterGen uses its neural network to gradually arrange these atoms into a stable crystal structure, much like turning random static (noise) into a recognizable image.​\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eProperty Conditioning: MatterGen can be given precise targets, such as high hardness, strong magnetism, or a specific chemical formula. The model continuously adjusts its output to meet these goals, generating materials tailored to specific requirements.​\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eLearning from Examples and Fine-Tuning: The tool was trained and is trained on a dataset of more than 600,000 real, experimentally confirmed crystal structures. For projects requiring niche or rare materials, a small labeled dataset can be provided, and MatterGen will adapt to suggest optimized candidates\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eSeveral examples have been reported showcasing MatterGen\u0026rsquo;s capabilities in materials development [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eTo create a super-hard material for cutting tools, MatterGen generated over 100 candidates with a bulk modulus exceeding 400 GPa, significantly harder than most common metals. In comparison, classical screening methods identified only 40 such candidates in existing databases.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eResearchers used MatterGen to design a new material targeting a bulk modulus of 200 GPa. After synthesis and laboratory testing, the measured hardness was 169 GPa, demonstrating how closely the model\u0026rsquo;s predictions aligned with real-world results.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eMatterGen can generate strong magnetic compounds by setting \u0026ldquo;high magnetic density\u0026rdquo; of 0.2 \u0026Aring;\u003csup\u003e\u0026minus;3\u003c/sup\u003e as a design goal. With just 180 density functional theory (DFT) property calculations, MatterGen is able to identify up to 18 stable, unique, and novel (SUN) structures exhibiting a magnetic density greater than 0.2 \u0026Aring;⁻\u0026sup3;.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe second AI tool from Microsoft, MatterSim, is a simulation engine powered by machine learning that predicts material behavior under real-world conditions with high speed and accuracy. It leverages \u003cem\u003edeep graph neural networks\u003c/em\u003e, uncertainty-aware sampling, and active learning. MatterSim functions as a zero-shot machine learning force field (MLFF), enabling end-to-end structure-to-property predictions, and operates across a wide range of elements, temperatures, and pressures. The main functional capabilities and core advantages of MatterSim are summarized as follows [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eDeep Learning Physics Engine: MatterSim learned from millions of physics-based calculations (\"first-principles\" data) covering thousands of different materials. It runs atomistic simulations: mimicking how a material\u0026rsquo;s atoms vibrate, move, and respond as conditions change.​\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eWide Range Coverage: It can predict thermodynamic, mechanical, and structural properties across the periodic table, for conditions from cold (0 K) to extremely hot (5000 K), and low to very high pressure (up to 1000 GPa).​\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eEfficiency and Accuracy: MatterSim provides results comparable to advanced quantum-mechanical methods, but within seconds instead of hours or days.​\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eRepresentative use cases of MatterSim demonstrate its versatility and predictive power:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003ePhase diagram computation: MatterSim can generate the phase diagram of a novel battery material within minutes, drastically reducing the time and experimental effort traditionally required.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003ePrediction of mechanical properties under extreme conditions: The tool enables accurate estimation of the mechanical strength of superconductors across a broad temperature range from \u0026minus;\u0026thinsp;200\u0026deg;C to +\u0026thinsp;1200\u0026deg;C.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eSimulation of lattice behavior: MatterSim can model how the crystal lattice of a newly developed magnetic material responds to external stimuli such as intense magnetic fields or thermal stress.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eMatterGen and MatterSim are designed to work synergistically. MatterGen can be used to generate promising new material candidates based on specified target properties. These candidates can then be evaluated using MatterSim, which rapidly predicts their stability, manufacturability, and functional performance under realistic conditions.\u003c/p\u003e \u003cp\u003eFor instance, when developing a material for solar panels that requires high electrical conductivity and thermal stability, MatterGen can generate suitable candidates that meet these criteria. MatterSim assesses thermal and mechanical stability over typical PV operating temperatures; optical and light\u0026ndash;matter effects are handled by separate methods (DFT, GW, Bethe\u0026ndash;Salpeter, device simulations).\u003c/p\u003e \u003cp\u003eAlthough both AI tools offer significant benefits, they also exhibit certain limitations and constraints:\u003c/p\u003e \u003cp\u003e\u003cul\u003e\u003cli\u003e\u003cp\u003eData Coverage: The accuracy of the models depends heavily on the availability and quality of training data. Very novel or exotic chemical systems may require additional fine-tuning or supplemental datasets.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eComplex Systems: While MatterSim is capable of handling a wide range of conditions, highly complex or multi-phase systems may exceed the current accuracy limits of neural network-based simulations.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eGeneralizability: The models are particularly effective for inorganic solid-state materials. However, support for organic compounds or hybrid interfaces remains limited at this stage.\u003c/p\u003e\u003c/li\u003e\u003cli\u003e\u003cp\u003eHardware Requirements: Optimal performance requires modern GPUs and compatible software libraries (e.g., CUDA, PyTorch). Deployment on certain hardware architectures, such as Apple Silicon, is still considered experimental.\u003c/p\u003e\u003c/li\u003e\u003c/ul\u003e\u003c/p\u003e \u003cp\u003ePlanned developments and future directions for Microsoft\u0026rsquo;s AI tools in materials science focus on expanding their capabilities and accessibility. Ongoing research and development aim to broaden the chemical space supported by the models, including more complex systems such as organic\u0026ndash;metal interfaces. Integration with experimental robotics and automated feedback loops is expected to enhance closed-loop \u0026ldquo;design-build-test\u0026rdquo; workflows, accelerating discovery cycles. Additionally, educational modules are being developed to make these tools more accessible for teaching purposes, supported by tailored curricula and datasets. Finally, MatterGen and MatterSim are poised to enable rapid hypothesis testing in emerging fields such as quantum materials, advanced semiconductors, and energy storage technologies.\u003c/p\u003e"},{"header":"3 Methodology","content":"\u003cp\u003eIn this study, the aim is to utilize open-source platforms for capstone projects in materials development, both in educational and research contexts [43, 44]. The long-term goal is to integrate MatterGen and MatterSim into materials science workflows to significantly accelerate the discovery and validation of new materials. For effective integration, the following tasks are necessary [38, 39]:\u003c/p\u003e\n\u003cp\u003e1.\u0026nbsp; \u0026nbsp;\u0026nbsp;Environment Setup and Installation\u003cbr\u003e\u0026nbsp;Installation followed the official GitHub documentation and arXiv paper. \u0026nbsp;MatterGen and MatterSim are available as open-source tools, implemented in Python and relying on PyTorch. Python 3.G Python 3.x (tested with version 3.9) was used. GPU support is strongly recommended for efficient computation. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e2.\u0026nbsp; \u0026nbsp;\u0026nbsp;Data Preparation\u003c/p\u003e\n\u003cp\u003eDatasets containing material structures and property labels should be curated. While MatterGen can operate without labels, performance improves with labeled data for conditional generation.\u003c/p\u003e\n\u003cp\u003e3.\u0026nbsp; \u0026nbsp;\u0026nbsp;Generative Design with MatterGen\u003c/p\u003e\n\u003cp\u003eNew material candidates can be generated using the pre-trained model. For targeted design, MatterGen may be fine-tuned on domain-specific datasets to guide generation toward desired compositions, symmetries, or property constraints, using approaches similar to ControlNet [45].\u003c/p\u003e\n\u003cp\u003e4.\u0026nbsp; \u0026nbsp;\u0026nbsp;High-Throughput Simulation with MatterSim\u003c/p\u003e\n\u003cp\u003eMatterSim is used to illustrate an educational validation workflow on Fe and Al as canonical test systems. The tool validates physical stability and predicts properties of candidates under a wide range of conditions.\u003c/p\u003e\n\u003cp\u003e5. Iterative Workflow: Flywheel Integration\u003cbr\u003eMatterGen and MatterSim can be integrated into an iterative loop: thousands of structural candidates are generated with MatterGen, simulated and triaged using MatterSim, and refined based on experimental or computational feedback. This closed-loop workflow accelerates the identification of promising materials and reduces the scope of experimental validation.\u003c/p\u003e\n\u003cp\u003e6.\u0026nbsp; \u0026nbsp;\u0026nbsp;Practical Applications\u003cbr\u003e\u0026nbsp;This workflow is applicable to a wide range of use cases requiring novel materials. Both academic and industrial research teams have adopted this approach for rapid prototyping and project-driven materials selection.\u003c/p\u003e\n\u003cp\u003e7.\u0026nbsp; \u0026nbsp;\u0026nbsp;Documentation, Community, and Support\u003cbr\u003e\u0026nbsp;Comprehensive usage guides and Application Programming Interface (API) documentation are available via the respective GitHub repositories and official documentation portals. Active participation in community discussions and Q\u0026amp;A forums is encouraged to gain support and insights from peer use cases.\u003c/p\u003e\n\u003cp\u003eThe short-term objective of this study is to establish a computational environment in which the tools can be installed, tested, and effectively applied, along with a functional workflow for simulation tasks. The following tasks have been completed to support this goal:\u003c/p\u003e\n\u003cp\u003e1. Installation and setup of MatterSim and MatterGen\u003c/p\u003e\n\u003cp\u003eIn addition to following official installation procedures, our experience revealed practical challenges specific to university teaching environments: Hardware Considerations: Student laptops and university workstations often have limited GPU memory (4-8 GB). This constrains supercell sizes for phonon calculations and trajectory lengths for molecular dynamics simulations. We recommend minimum 8 GB GPU memory for educational use, with 16 GB preferred for advanced projects. Software Environment Issues: On Windows systems, Conda/Visual C++ Build Tools compatibility is critical. Students encountered issues with CUDA vs MPS (Apple Silicon) platforms, with numerical discrepancies in some calculations. Switching to CPU execution or CUDA-enabled systems resolved these issues. Realistic Simulation Scope: For semester-long projects, structural relaxation and phonon analysis are highly reliable. Molecular dynamics simulations are feasible for small systems. However, automated phase diagram generation requires additional validation and is best suited for advanced projects with instructor support. Timeline Expectations: Installation and environment setup typically require 2-4 hours. Initial validation simulations (Fe, Al) take 1-2 weeks. Full capstone projects span 8-12 weeks.\u003c/p\u003e\n\u003cp\u003eTo enable the use of MatterSim and MatterGen, a clean and isolated Python environment should be created. It is recommended to use Conda or Mamba to avoid version conflicts and to ensure proper integration of Visual C++ tools on Windows systems. The following procedures are necessary for implementation:\u003c/p\u003e\n\u003cp\u003ea) Software requirements\u003c/p\u003e\n\u003cp\u003e- Windows 10/11 with administrator rights\u003c/p\u003e\n\u003cp\u003e- Python 3.G (MatterSim recommends version 3.6 for maximum stability)\u003c/p\u003e\n\u003cp\u003e- Conda or Mamba installed (instructions: https://docs.conda.io/)\u003c/p\u003e\n\u003cp\u003e- Visual C++ Build Tools (for Windows compilers)\u003c/p\u003e\n\u003cp\u003eb) Create virtual environment\u003c/p\u003e\n\u003cp\u003ec) Update Pip and build helper packages\u003c/p\u003e\n\u003cp\u003ed) Install MatterSim\u003c/p\u003e\n\u003cp\u003e- Stable version from PyPI\u003c/p\u003e\n\u003cp\u003e- Developer version directly from GitHub\u003c/p\u003e\n\u003cp\u003e- Installation from cloned source code\u003c/p\u003e\n\u003cp\u003ee) Install MatterGen\u003c/p\u003e\n\u003cp\u003ef) Final test\u003c/p\u003e\n\u003cp\u003e2. Code integration and workflows\u003c/p\u003e\n\u003cp\u003eA Python-based graphical user interface (GUI) application was developed to facilitate atomistic simulations. The application is based on the MatterSim Machine Learning Force Field (MLFF) model, version v1.0.0-1M, and integrates several Python libraries, including the Atomic Simulation Environment (ASE), Phonopy for phonon calculations, and Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) for molecular dynamics simulations.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe GUI is not simply a thin wrapper around existing libraries, but a didactic layer that encodes good-practice workflows for atomistic simulations in educational settings. The architecture consists of: Front-end: Built using standard Python widget frameworks (Tkinter/PyQt) with integrated plotting libraries (Matplotlib) for real-time visualization. The layout uses progressive disclosure, showing basic parameters by default with expandable panels for advanced options. Back-end: Orchestrates ASE calculator instantiation, MatterSim force-field loading, LAMMPS interface configuration, and data pipelines between components. The GUI manages supercell generation, displacement amplitude calculations for phonon analysis, ensemble choices (NVT/NPT) for molecular dynamics, and result formatting. Workflow Encoding: The GUI embeds validated workflows tested in our educational context: sensible defaults for displacement amplitudes (0.01 Å), supercell sizes (2×2×2 for metals), temperature ranges (0-1500 K), and timesteps (1 fs for MD). Human-Centric Design Principles are:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(a) Sensible defaults to minimize decision paralysis for novice users\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(b) Inline documentation for each parameter with tooltips explaining physical meaning\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(c) Progressive disclosure of advanced options to avoid overwhelming beginners\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e(d) Workflow templates for common tasks (structural relaxation, phonon calculation). The GUI is open-source and extensible, enabling instructors and students to adapt it to new materials and future versions of MatterSim and MatterGen. New figures will be added showing: (1) annotated screenshot with key GUI modules and workflows, (2) flowchart illustrating how the GUI orchestrates MatterSim, ASE, Phonopy, and LAMMPS.\u003c/p\u003e\n\u003cp\u003eThe GUI enables users to select crystal structures, perform computational tasks, and analyze simulation results. The application provides four main functional modules:\u003c/p\u003e\n\u003cp\u003e- Structural Relaxation: Optimization of atomic positions within crystal structures to minimize total energy and achieve equilibrium geometry.\u003c/p\u003e\n\u003cp\u003e- Phonon Calculations: Determination of phonon dispersion relations and the phonon density of states, which are essential for understanding thermal and vibrational properties.\u003c/p\u003e\n\u003cp\u003e- Molecular Dynamics Simulations: Execution of simulations in canonical (NVT: constant number of particles, volume, and temperature) and isothermal-isobaric (NPT: constant number of particles, pressure, and temperature) ensembles, including temperature ramping and pressure variation studies.\u003c/p\u003e\n\u003cp\u003e- Thermodynamic Property Estimation: Calculation of macroscopic quantities such as Gibbs free energy, entropy, and heat capacity based on atomistic simulation data.\u003c/p\u003e\n\u003cp\u003eThe first phase of the project focused on using MatterSim to simulate material properties, such as those of Fe, Al, and their alloys. The core functionalities, e.g. structural relaxation, force and stress calculations, and phonon analysis, were successfully executed during initial testing. In contrast, more advanced thermodynamic calculations were only partially supported in the open-source version of MatterSim utilized in this study. All simulations were carried out as planned, and the resulting data were systematically compared with values reported in the scientific literature.\u003c/p\u003e"},{"header":"4 Results and Discussion","content":"\u003cp\u003eThe most important findings of this study can be summarized as follows:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eThe developed Python-based Graphical User Interface (GUI) enables intuitive execution of simulations with MatterSim and enhances the transparency of the underlying workflows for users.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe essential computational tasks \u0026ndash; structural relaxation, force and stress calculations, and phonon analysis \u0026ndash; were successfully implemented. The results qualitatively align with expectations reported in the literature.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eMolecular dynamics simulations were feasible in selected cases; however, limitations were observed in the automated phase analysis functionality.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThermodynamic properties could only be estimated to a limited extent due to constraints in the open-source version of MatterSim used.\u003c/p\u003e\n \u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe detailed findings will be presented and discussed below.\u003c/p\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e4.1 Python-based Graphical User Interface (GUI)\u003c/h2\u003e\n \u003cp\u003eA Python-based Graphical User Interface (GUI) was successfully developed to enable convenient use and execution of simulations. Parameters such as chemical element, lattice structure, lattice constant, temperature, and pressure can be entered. Simulation modes, including bulk relaxation, bulk phonon, molecule relaxation, and molecule phonon, can be executed. Various material properties, such as total energy, atomic forces and stress, heat capacity, and vapor pressure as a function of temperature, can be calculated. The developed GUI for visualization and simulation is demonstrated using examples of Fe (body-centered cubic structure, bcc) and Al (face-centered cubic structure, fcc), as shown in Fig. \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\n \u003ch2\u003e4.2 Relaxation and force/stress calculation\u003c/h2\u003e\n \u003cp\u003eThe relaxation function enables the optimization of crystal structures using MatterSim [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. In this process, the MatterSim calculator is linked to an atomic structure object provided by the Atomic Simulation Environment (ASE). The structure is then relaxed using an optimization algorithm, such as the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method. As a result, the parameters \u0026ndash; total energy, atomic forces, and stress tensors \u0026ndash; were calculated and are presented in Fig. \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e for Fe (bcc structure) and Al (fcc structure), for example, at a pressure of 1 GPa and a room temperature of 300 K:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eThe total energy E (in eV) represents the lattice potential energy modeled by MatterSim, which is the sum of all interatomic interactions. It reflects the energetic state of the atomic configuration and is fundamental for understanding the material\u0026rsquo;s stability and chemical properties [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. In general, a lower total energy indicates a more stable atomic configuration, such as a crystal unit cell, a molecule, or an atomic cluster, where an energy minimum corresponds to a relaxed and stable structure. This work reports total energies of \u0026minus;\u0026thinsp;8.46 eV for idealized Fe (bcc structure) and \u0026minus;\u0026thinsp;3.72 eV for Al (fcc structure). Since the bcc Fe unit cell contains two atoms, the total energy of \u0026minus;\u0026thinsp;8.46 eV corresponds to approximately \u0026minus;\u0026thinsp;4.23 eV per atom. For fcc Al, the reported total energy already refers to a single atom. Therefore, the total energy normalized per atom for Fe (bcc) and Al (fcc) in this study is \u0026minus;\u0026thinsp;4.23 eV and \u0026minus;\u0026thinsp;3.72 eV, respectively. These values are in good agreement with the cohesive energies reported in the literature [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e], which are defined as the energy required to separate neutral atoms in their ground electronic state from the solid at 0 K and 1 atm, and amount to \u0026minus;\u0026thinsp;4.28 eV for Fe and \u0026minus;\u0026thinsp;3.39 eV for Al.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eAtomic forces F (in eV/\u0026Aring;) are obtained by computing the gradient of the energy with respect to atomic positions via automatic differentiation [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. These forces are essential for structural optimization and molecular dynamics simulations. The force vector (F = [Fx, Fy, Fz]) with a 1\u0026times;3 output represents the net force acting on the considered atoms. Values close to 0 eV/\u0026Aring; indicate that the system is in force equilibrium. For relaxed metallic structures, residual forces typically range between 10⁻\u0026sup3; and 10⁻⁵ eV/\u0026Aring; [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. The computed atomic forces for Fe and Al in this study are 0 eV/\u0026Aring;, confirming that the system is in force equilibrium.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe stress tensor \u0026sigma; (in eV/\u0026Aring; or GPa) characterizes mechanical tensions within the material and is relevant for determining mechanical properties. The 3\u0026times;3 matrix describes the stress tensor \u0026sigma; [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]:\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cdiv id=\"Equ1\" class=\"Equation\"\u003e\n \u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equ1\" name=\"EquationSource\"\u003e$$\\:\\sigma\\:=\\left[\\begin{array}{ccc}{\\sigma\\:}_{xx}\u0026amp;\\:{\\sigma\\:}_{xy}\u0026amp;\\:{\\sigma\\:}_{xz}\\\\\\:{\\sigma\\:}_{yx}\u0026amp;\\:{\\sigma\\:}_{yy}\u0026amp;\\:{\\sigma\\:}_{yz}\\\\\\:{\\sigma\\:}_{zx}\u0026amp;\\:{\\sigma\\:}_{zy}\u0026amp;\\:{\\sigma\\:}_{zz}\\end{array}\\right]$$\u003c/div\u003e\n \u003cdiv class=\"EquationNumber\"\u003e1\u003c/div\u003e\n \u003c/div\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003eDiagonal elements (\u0026sigma;xx, \u0026sigma;yy, \u0026sigma;zz) represent normal stresses, while off-diagonal elements (\u0026sigma;xy, \u0026sigma;xz, \u0026sigma;yz \u0026hellip;) represent shear stresses. In an ideal relaxed metal crystal without external loads, all components should be close to 0 GPa. This was confirmed by the results of this study, which illustrate both the structural stability and the predictive accuracy of the MatterSim model.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e4.3 Phonon calculations\u003c/h2\u003e\n \u003cp\u003ePhonon calculations were performed using MatterSim in combination with Phonopy via the frozen-phonon method [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Phonopy generates supercells with small atomic displacements (typically\u0026thinsp;~\u0026thinsp;0.01 \u0026Aring;), and MatterSim computes the corresponding atomic forces using its M3GNet-based machine-learning potential. These forces are then processed by Phonopy to construct the force-constant matrix, from which phonon dispersion relations, phonon density of states (DOS), and dynamical stability are derived. Rigorous benchmarking of MatterSim across thousands of materials has been reported in the official technical documentation and MatBench Discovery platform. Our results on Fe and Al demonstrate how this validated model can be translated into a student-friendly workflow. This approach enables efficient and accurate prediction of vibrational properties critical for thermal conductivity, heat capacity, and material stability.\u003c/p\u003e\n \u003cp\u003eThis method demonstrated strong performance in the tests, with the phonon DOS and dispersion spectra for Fe and Al presented in Fig. \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e. The phonon dispersions and densities generated in the GUI exhibit consistent behavior, confirming the physical validity of the results: no imaginary frequencies occur, indicating stable lattice structures; the acoustic and optical branches show the expected shapes; and the spectra qualitatively agree with known DFT reference data [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]. Comparison with published DFT results shows good agreement in phonon frequencies for Fe and Al and supports the use of MatterSim for thermal-property predictions in educational workflows. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e compares key phonon properties between MatterSim predictions and published DFT references.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eQuantitative comparison of phonon properties: MatterSim versus DFT\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003ccolgroup cols=\"4\"\u003e\u003c/colgroup\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" colname=\"c1\"\u003e\n \u003cp\u003eMaterial\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eProperties\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003eMatterSim\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003eDFT Reference\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\n \u003cp\u003eFe (bcc)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eMax frequency (THz)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e8.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e8.4 [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDebye temperature (K)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e455\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e470 [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eAcoustic slope (km/s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e6.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e6.3 [\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e\n \u003cp\u003eAl (fcc)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eMax frequency (THz)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e9.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e9.5 [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eDebye temperature (K)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e425\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e428 [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" colname=\"c2\"\u003e\n \u003cp\u003eAcoustic slope (km/s)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c3\"\u003e\n \u003cp\u003e6.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" colname=\"c4\"\u003e\n \u003cp\u003e6.5 [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003eThis high accuracy stems from MatterSim\u0026rsquo;s training on numerous distorted supercells, which enhances its ability to predict forces for phononic displacements. Consequently, phonon calculations rank among the most reliable features of the open-source MatterSim version.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e4.4 Molecular dynamics (MD)\u003c/h2\u003e\n \u003cp\u003eIn the Molecular Dynamics (MD) section of the Graphical User Interface (GUI), the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) acts as the integration engine, while MatterSim provides the interatomic forces [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. This setup enables simulations in the canonical ensemble (NVT) and the isothermal-isobaric ensemble (NPT), such as temperature ramping or pressure variation runs. In practice, expected physical trends were observed; for instance, the internal energy increased with rising temperature. However, generating a fully automated phase diagram was not possible because the version of MatterSim used does not support this functionality. Nevertheless, the raw data from the graphical user interface can be utilized to plot temperature and energy curves that qualitatively indicate phase transitions.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e4.5 Thermodynamic properties\u003c/h2\u003e\n \u003cp\u003eThe Graphical User Interface (GUI) includes experimental features for evaluating thermodynamic properties such as free energy and heat capacity (Fig. \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e). These calculations are based either on the phonon density of states or on results from molecular dynamics simulations, and are used to estimate quantities such as heat capacity at constant pressure (\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{p}\\)\u003c/span\u003e\u003c/span\u003e) and enthalpy [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. The heat capacity curves \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{p}\\left(T\\right)\\:\\)\u003c/span\u003e\u003c/span\u003egenerated by the GUI exhibit trends that are partially plausible but also include some inconsistencies. Typically, the specific heat capacity \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{p}\\:\\)\u003c/span\u003e\u003c/span\u003eshows a sharp increase (or discontinuity) at the melting point due to the solid-to-liquid phase transition, which is confirmed by calculations for Fe and Al in this study. However, the \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:{C}_{p}\\:\\)\u003c/span\u003e\u003c/span\u003ecurves are only partially reliable and should be reported as a known limitation. These deviations arise from numerical issues in the GUI (due to differentiation of unsmoothed free-energy curves), limitations of MatterSim (missing electronic contributions and incomplete free-energy data), and noise introduced by small supercells in phonon DOS calculations.\u003c/p\u003e\n \u003cp\u003eThe calculated vapor pressure curves show the expected trend of increasing with higher temperatures. However, the absolute values should be treated with caution, which is not surprising given the limitations of the model. Determining vapor pressure requires accurate enthalpies of vaporization, realistic liquid and gas phases, and precise free energies [\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. MatterSim cannot provide these because it was not trained for gas phases, does not reliably represent liquids, and the open version only delivers approximate static energies. Additionally, the GUI approach, such as differentiation or Molecular Dynamics-based volume changes, does not allow a valid determination of vapor pressure. Therefore, vapor pressure cannot be accurately computed using the open-source MatterSim version.\u003c/p\u003e\n \u003cp\u003eMoreover, due to limitations in the open-source version of MatterSim used, comprehensive thermodynamic analyses were not fully supported. Consequently, only partial results could be obtained. For example, the calculated phase diagram of Fe (Fig. \u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e) indicated tendencies of possible phase transitions, such as from the fcc structure to the bcc structure. Nevertheless, a complete determination of the phase diagram was not feasible. The main reasons include incomplete Gibbs free energy data (electronic, configurational, and high-temperature contributions are missing), lack of Application Programming Interface (API) support (since the open-source MatterSim version does not provide integrated phase diagram functionality), and a highly simplified data basis with only a few temperature and pressure points (resulting in irregular and unstable diagrams). Despite these limitations, the calculated phase diagram of Fe is partly consistent with reference data (Fig. \u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e) [\u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]. Furthermore, attempts to create phase diagrams for Fe\u0026ndash;C and Al\u0026ndash;Si alloys were unsuccessful. Only eutectic points (Al\u0026ndash;Si at 577\u0026deg;C and Fe\u0026ndash;C at 1147\u0026deg;C) as well as the eutectoid point of Fe\u0026ndash;C at 723\u0026deg;C could be confirmed in this study. Nevertheless, complete phase diagrams for Fe\u0026ndash;C and Al\u0026ndash;Si alloys could not be generated; therefore, the phase diagrams of these alloys are not reported here.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e4.5 Limitations and Challenges\u003c/h2\u003e\n \u003cp\u003eThe practical implementation of the simulations revealed three key limitations that should be considered in future work involving MatterSim:\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e1. Computational and Storage Resources: Large supercells, extended molecular dynamics simulations, and especially the training or fine-tuning of large-scale models demand substantial Graphics Processing Unit (GPU) memory. In the absence of sufficient hardware resources, limitations in batch size, supercell dimensions, and model complexity can reduce the significance and reliability of certain results.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e2. Infrastructure and Platform Dependency: Managing large model weights and trajectory datasets requires robust data transfer and version control mechanisms. Unstable network connections or the absence of a Large File Storage (LFS) infrastructure can delay workflows. Additionally, numerical discrepancies may arise between computing platforms (e.g., Metal Performance Shaders (MPS) on Apple Silicon versus Compute Unified Device Architecture (CUDA) on x86 systems). In problematic cases, switching to Central Processing Unit (CPU) execution or alternative hardware platforms may be necessary.\u003c/p\u003e\n \u003c/span\u003e \u003cspan\u003e\n \u003cp\u003e3. Methodological Limitations and Validation: The publicly available version of MatterSim used in this project supports reliable structural relaxations, force calculations, and phonon analyses. However, it does not provide full support for automated and comprehensive thermodynamic workflows, such as the generation of complete phase diagrams. Machine learning-based force fields are inherently data-driven and may exhibit systematic deviations when applied outside their training domain. Therefore, targeted validation using Density Functional Theory (DFT) or experimental data is essential. In this project, selected predictions were cross-validated using DFT, with deviations typically within the single-digit percentage range.\u003c/p\u003e\n \u003c/span\u003e\n \u003cp\u003eMatterGen was not used productively in this phase of the project. For generative applications, targeted training or fine-tuning on appropriate datasets and the development of a validation pipeline are still required.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"5 Conclusions and Outlook","content":"\u003cp\u003eThis study explores the integration of the open-source platforms MatterGen and MatterSim into a computational workflow for materials development in educational and research settings. A Python-based Graphical User Interface (GUI) was developed to support key simulation tasks, including structural relaxation, phonon analysis, molecular dynamics, and thermodynamic property estimation.\u003c/p\u003e \u003cp\u003ePhonon spectra generated using MatterSim in combination with Phonopy showed good agreement with Density Functional Theory (DFT) reference data. Molecular dynamics simulations, powered by the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS), revealed expected physical trends, although automated phase diagram generation was not supported. Thermodynamic properties such as free energy and heat capacity were partially estimated, with qualitative trends successfully reproduced.\u003c/p\u003e \u003cp\u003eChallenges encountered included limitations in hardware resources (especially Graphics Processing Unit memory), infrastructure dependencies (such as the need for Large File Storage systems), and methodological constraints in the open-source version of MatterSim. While MatterGen was not actively used in this phase, it shows strong potential for future generative design tasks.\u003c/p\u003e \u003cp\u003eShort-term recommendations include leveraging GPU cluster resources, ensuring stable data infrastructure (e.g., Large File Storage), integrating free energy methods such as the Quasi-Harmonic Approximation and thermodynamic integration, and validating results systematically using Density Functional Theory. Long-term potential lies in combining generative models with machine learning-based force fields to enable efficient materials screening workflows. Future directions involve exploring MatterGen and merging generative and predictive models for rapid screening. This work highlights AI\u0026rsquo;s role as a valuable resource for both teaching materials engineering and accelerating computational materials discovery. The custom Python-based GUI, integrated with MatterSim, is available as open source on GitHub for public access and further development (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git\u003c/span\u003e\u003cspan address=\"https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e"},{"header":"Abbreviations ","content":"\u003cp\u003eAI\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eASE\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Atomic Simulation Environment\u003c/p\u003e\n\u003cp\u003eAPI\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Application Programming Interface\u003c/p\u003e\n\u003cp\u003eBFGS\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Broyden–Fletcher–Goldfarb–Shanno\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCNN\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Convolutional Neural Network\u003c/p\u003e\n\u003cp\u003eDFT\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Density Functional Theory\u003c/p\u003e\n\u003cp\u003eDOS\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Density of States\u003c/p\u003e\n\u003cp\u003eFEM\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Finite Element Method\u003c/p\u003e\n\u003cp\u003eFCC\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Face-Centered Cubic\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBCC\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Body-Centered Cubic\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGPU\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Graphics Processing Unit\u003c/p\u003e\n\u003cp\u003eGUI\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Graphical User Interface\u003c/p\u003e\n\u003cp\u003eLAMMPS\u0026nbsp; \u0026nbsp; \u0026nbsp;Large-scale Atomic/Molecular Massively Parallel Simulator\u003c/p\u003e\n\u003cp\u003eLFS\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Large File Storage\u003c/p\u003e\n\u003cp\u003eML\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Machine Learning\u003c/p\u003e\n\u003cp\u003eMLFF\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Machine Learning Force Field\u003c/p\u003e\n\u003cp\u003eMPS\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Metal Performance Shaders\u003c/p\u003e\n\u003cp\u003eNPT\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Isothermal-Isobaric Ensemble (constant Number of particles, Pressure, Temperature)\u003c/p\u003e\n\u003cp\u003eNVT\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Canonical Ensemble (constant Number of particles, Volume, Temperature)\u003c/p\u003e\n\u003cp\u003ePV\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Photovoltaic\u003c/p\u003e\n\u003cp\u003eR\u0026amp;D\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;Research and Development\u003c/p\u003e\n\u003cp\u003eSUN\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;Stable, Unique, and Novel\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e This manuscript is an extended version of the paper \u003cem\u003e\u0026lsquo;Artificial Intelligence in Materials Science Education and Research \u0026amp; Development\u0026rsquo;\u003c/em\u003e, originally presented at \u003cem\u003eThe 2025 International Scientific Conference on Media Education in the age of AI and Digital Transformation\u0026nbsp;\u003c/em\u003eon 25 December 2025 in Hanoi, Vietnam. The authors would like to thank the organizers of the conference for providing the opportunity to present the preliminary version of this work and for granting permission to reuse the content of the published conference paper for submission to another journal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e The authors confirm their contributions to the paper as follows: Conceptualization and methodology: Gia Khanh Pham, Ulrick Edwin Nguegang Tsopmo, Marta Segura Cubells. Software: Ulrick Edwin Nguegang Tsopmo. Data collection and analysis: Ulrick Edwin Nguegang Tsopmo, Gia Khanh Pham. Draft manuscript: Gia Khanh Pham, Ulrick Edwin Nguegang Tsopmo. Manuscript revisions: Ulrick Edwin Nguegang Tsopmo, Marta Segura Cubells. Supervision: Gia Khanh Pham. All authors have read and approved the final version of the manuscript for publication.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e No funding was received for conducting this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData Availability\u003c/strong\u003e The data from this study and the open-source version of the GUI are available on GitHub (https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git) or can be obtained from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of interest\u003c/strong\u003e The authors declare no conflicts of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics Approval\u0026nbsp;\u003c/strong\u003eThis manuscript is an extended version of a conference paper. The authors confirm that this work complies with the ethical responsibilities and standards for research and publication.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eOpen Access\u003c/strong\u003e This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article\u0026apos;s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article\u0026apos;s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eNematov, D., \u0026amp; Hojamberdiev, M. (2025). Machine learning\u0026ndash;driven materials discovery: Unlocking next-generation functional materials \u0026ndash; A review. Computational Condensed Matter, 45 (2025), 1-23. https://doi.org/10.1016/j.cocom.2025.e01139\u003c/li\u003e\n \u003cli\u003eYılmaz, \u0026Ouml;. (2024). Personalised learning and artificial intelligence in science education: Current state and future perspectives. Educational Technology Quarterly, 2024(3), 255\u0026ndash;274. https://doi.org/10.55056/etq.744\u003c/li\u003e\n \u003cli\u003eGuinan, G., Salvador, A., Smeaton, M. A., Glaws, A., Egan, H., Wyatt, B. C., Anasori, B., Fiedler, K. R., Olszta, M. J.,\u0026amp; Spurgeon, S. R. (2025). Mind the gap: Bridging the divide between AI aspirations and the reality of autonomous characterization. APL Mach. Learn, 3(2), 020903. https://doi.org/10.1063/5.0267699\u003c/li\u003e\n \u003cli\u003eTom, G., Schmid, S. P., Baird, S. G., Cao, Y., Darvish, K., Hao, H., Lo, S., Pablo-Garc\u0026iacute;a, S., Rajaonson, E. M., Skreta, M., Yoshikawa, N., Corapi, S., Akkoc, G. D., Strieth-Kalthoff F., Seifrid, M., \u0026amp; Aspuru-Guzik, A. (2024). Self-driving laboratories for chemistry and materials science. Chemical Reviews, 124, 9633\u0026minus;9732. https://doi.org/10.1021/acs.chemrev.4c00055\u003c/li\u003e\n \u003cli\u003eMa, J., Cao, B., Dong, S., Tian Y., Wang M., Xiong J., \u0026amp; Sun S. (2024). MLMD: A programming-free AI platform to predict and design novel materials. npj Computational Materials, 10 (59). https://doi.org/10.1038/s41524-024-01243-4\u003c/li\u003e\n \u003cli\u003eGu, Z., Tran, H., Manjunath, A., Chen, D., \u0026amp; Fu, S. (2024). AI automated microstructure analysis for intelligent manufacturing. Open Journal of Robotics and Automation Technology, 3(2), 110\u0026ndash;122. https://irispublishers.com/ojrat/pdf/OJRAT.MS.ID.000533.pdf\u003c/li\u003e\n \u003cli\u003eNa, J., Kim, S. J., Kim, H., Kang, S. H., \u0026amp; Lee, S. (2023). A unified microstructure segmentation approach via human-in-the-loop machine learning. Acta Materialia, 255. https://doi.org/10.1016/j.actamat.2023.119086\u003c/li\u003e\n \u003cli\u003ePham, G. K., Yalcin, K., Binti Wan, A. W. A. N., Nguyen, A. S., \u0026amp; Trinh, V. T. (2024). Development of an Open-Source Software Tool for Microstructure Analysis of Materials Using Artificial Intelligence. Key Engineering Materials, 1004, 103\u0026ndash;110. https://doi.org/10.4028/p-vcog1h\u003c/li\u003e\n \u003cli\u003eKabudi, T., Pappas, I., \u0026amp; Olsen, D. H. (2021). AI-enabled adaptive learning systems: A systematic mapping of the literature. Computers and Education: Artificial Intelligence, 2. https://doi.org/10.1016/j.caeai.2021.100017\u003c/li\u003e\n \u003cli\u003eKatona, J., \u0026amp; Gyonyoru, K. I. K. (2025). Integrating AI-based adaptive learning into the flipped classroom model to enhance engagement and learning outcomes. Computers and Education: Artificial Intelligence, 8. https://doi.org/10.1016/j.caeai.2025.100392\u003c/li\u003e\n \u003cli\u003eeLearning Industry. (2024). Adaptive learning with AI: Revolutionizing personalized education. eLearning Industry. https://elearningindustry.com/understanding-adaptive-learning-how-ai-is-revolutionizing-personalized-education (accessed on 20.10.2025)\u003c/li\u003e\n \u003cli\u003eLiu, Y., Yang, Z., Yu, Z., Liu, Z., Liu, D., Lin, H., Li, M., Ma, S., Avdeev, M., Shi, S. (2023). Generative artificial intelligence and its applications in materials science: Current situation and future perspectives. Journal of Materiomics, 9(4), 798-816. https://doi.org/10.1016/j.jmat.2023.05.001\u003c/li\u003e\n \u003cli\u003eKumar, K. (2025). Generative AI in materials science: Accelerating discovery through inverse design. SSRN Paper No. 5251184. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5251184\u003c/li\u003e\n \u003cli\u003eErduran, S., \u0026amp; Levrini, O, (2024). The impact of artificial intelligence on scientific practices: an emergent area of research for science education. International Journal of Science Education, 46(18), 1982\u0026ndash;1989. https://doi.org/10.1080/09500693.2024.2306604\u003c/li\u003e\n \u003cli\u003eDeCost, B. L., Hattrick-Simpers, J. R., Trautt, Z., Kusne, A. G., Campo, E., \u0026amp; Green, M. L. (2020). Scientific AI in materials science: A path to a sustainable and scalable paradigm. Mach. Learn.: Sci. Technol. 1, 033001. https://doi.org/10.1088/2632-2153/ab9a20\u003c/li\u003e\n \u003cli\u003eBatool, A., Sanumi, O., \u0026amp; Jankovic, J. (2024). Application of artificial intelligence in the materials science, with a special focus on fuel cells and electrolyzers. Energy and AI, 18. https://doi.org/10.1016/j.egyai.2024.100424\u003c/li\u003e\n \u003cli\u003ePeivaste, I., Belouettar, S., Mercuri, F., Fantuzzi, N., Dehghani, H., Izadi, R., Ibrahim, H., Lengiewicz, J., Belouettar-Mathis, M., Bendine, K., Makradi, A., Horsch, M., Klein, P., El Hachemi, M., Preisig, H. A., Rezgui, Y., Konchakova, N., \u0026amp; Daouadji, A. (2025). Artificial intelligence in materials science and engineering: Current landscape, key challenges, and future trajectories. Composite Structures, 372, 119419. https://doi.org/10.1016/j.compstruct.2025.119419\u003c/li\u003e\n \u003cli\u003eBai, X., \u0026amp; Zhang, X. (2025). Artificial Intelligence-Powered Materials Science. Nano-Micro Lett. 17, 135. https://doi.org/10.1007/s40820-024-01634-8\u003c/li\u003e\n \u003cli\u003eChitturi, S. R., Ramdas, A., Wu, Y., Rohr, B., Ermon, S., Dionne, J., da Jornada, F. H., Dunne, M., Tassone, C., Neiswanger, W., \u0026amp; Ratner, D. (2024). Targeted materials discovery using Bayesian algorithm execution. npj Computational Materials, 10, 156. https://doi.org/10.1038/s41524-024-01326-2\u003c/li\u003e\n \u003cli\u003eMerchant, A., Batzner, S., Schoenholz, S.S., Aykol, M., Cheon, G., Cubuk, E. D. (2023). Scaling deep learning for materials discovery. Nature, 624, 80\u0026ndash;85. https://doi.org/10.1038/s41586-023-06735-9\u003c/li\u003e\n \u003cli\u003ePriyadarshini, M. S., Romiluyi, O., Wang, Y., Miskin, K., Ganley, C., \u0026amp; Clancy, P. (2024). PAL 2.0: A physics driven Bayesian optimization framework for material discovery. Materials Horizons, 11(28), 781\u0026ndash;791. https://doi.org/10.1039/d3mh01474f\u003c/li\u003e\n \u003cli\u003eAlrfou, K., Kordijazi, A., \u0026amp; Zhao, T. (2022). Computer vision methods for the microstructural analysis of materials: The State-of-the-art and Future Perspectives. arXiv:2208.04149. https://doi.org/10.48550/arXiv.2208.04149\u003c/li\u003e\n \u003cli\u003eMishra, S. P., \u0026amp; Rahul, M. R. (2023). A detailed study of convolutional neural networks for the identification of microstructure. Materials Chemistry and Physics, 308, 128275. https://doi.org/10.1016/j.matchemphys.2023.128275\u003c/li\u003e\n \u003cli\u003eGao, S., Huang, Q., Huang, C., Li, C., Liu, K., Sa, B., Yu, Y., Xue, D., Liu, Z., \u0026amp; Dai, M. (2025). Deep generative model for the inverse design of Van der Waals heterostructures. Sci Rep 15, 23023. https://doi.org/10.1038/s41598-025-06432-9\u003c/li\u003e\n \u003cli\u003eQin, C., Liu, J., Ma, S., Du, J., Jiang, G., \u0026amp; Zhao, L. (2024). Inverse design of semiconductor materials with deep generative models. Journals of Materials Chemistry A, 12, 22689-22702. https://doi.org/10.1039/d4ta02872d\u003c/li\u003e\n \u003cli\u003ePapadimitriou, I., Gialampoukidis, I., Vrochidis, S., \u0026amp; Kompatsiaris, I. (2024). AI methods in materials design, discovery and manufacturing: A review. Computational Materials Science, 235, 112793. https://doi.org/10.1016/j.commatsci.2024.112793\u003c/li\u003e\n \u003cli\u003eXu, P., Ma, Y., Lu, W., Li, M., Zhao, W., \u0026amp; Dai, Z. (2025). Multi-objective optimization in machine learning assisted materials design and discovery. J. Mater. Inf. 2025, 5, 26. http://dx.doi.org/10.20517/jmi.2024.108\u003c/li\u003e\n \u003cli\u003eJiang, X., Xue, D., Bai, Y., Wang, W. Y., Liu, J., Yang, M., \u0026amp; Su, Y. (2025). AI4Materials: Transforming the landscape of materials science and engineering. Review of Materials Research, 1(1), 100010. https://doi.org/10.1016/j.revmat.2025.100010\u003c/li\u003e\n \u003cli\u003eMerchant, A., \u0026amp; Cubuk, E. D. (2023). Millions of new materials discovered with deep learning. Google DeepMind. https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/ (accessed on 20.10.2025)\u003c/li\u003e\n \u003cli\u003eSzymanski, N. J., Rendy, B., Fei, Y., Kumar, R. E., He, T., Milsted, D., McDermott, M. J., Gallant, M., Cubuk, E. D., Merchant, A., Kim, H., Jain, A., Bartel, C. J., Persson, K., Zeng, Y., \u0026amp; Ceder, G. (2023). An autonomous laboratory for the accelerated synthesis of novel materials. Nature 624, 86\u0026ndash;91. https://doi.org/10.1038/s41586-023-06734-w\u003c/li\u003e\n \u003cli\u003eMianroodi, J.R., H. Siboni, N. \u0026amp; Raabe, D. (2021). Teaching solid mechanics to artificial intelligence \u0026ndash; a fast solver for heterogeneous materials. npj Comput Mater 7, 99. https://doi.org/10.1038/s41524-021-00571-z\u003c/li\u003e\n \u003cli\u003eTrewartha, A., \u0026amp; Torrisi, S. (2024). Toyota Research Institute to Launch Challenge to Accelerate Research in New Advanced Materials. Toyota Research Institute. https://medium.com/toyotaresearch/toyota-research-institute-to-launch-program-to-accelerate-research-in-new-advanced-materials-d52d5d65e8d9 (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eMulholland, G. J. (2018). AI for accelerated materials discovery. Citrine. https://worldmaterialsforum.com/files/Presentations2018/PS5/WMF%202018%20Citrine%20Informatics%20Mulholland.pdf (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eFraunhofer IWM (2025). KI-Konzepte f\u0026uuml;r Materialentwicklung und Werkstofftechnologie. https://www.iwm.fraunhofer.de/de/geschaeftsfelder/bauteilsicherheit-und-leichtbau/meso-und-mikromechanik/KI-konzepte_materialentwicklung_werkstofftechnologie.html (accessed on 20.10.2025)\u003c/li\u003e\n \u003cli\u003eDFKI Deutsches Forschungszentrum f\u0026uuml;r K\u0026uuml;nstliche Intelligenz (2022). Machine Learning f\u0026uuml;r die Herstellungsprozess-Simulation f\u0026uuml;r Verbundwerkstoffe. https://www.dfki.de/web/news/machine-learning-simulation-verbundwerkstoffe (accessed on 20.10.2025)\u003c/li\u003e\n \u003cli\u003eWu, J., Torresi, L., Hu, M., Reiser, P., Zhang, J., Rocha-Ortiz, J. S., Wang, L., Xie, Z., Zhang, K., Park, B., Barabash, A., Zhao, Y., Luo, J., Wang, Y., L\u0026uuml;er, L., Deng, L.-L., Hauch, J. A., Guldi, D. M., P\u0026eacute;rez-Ojeda, M. E., Seok, S. I., Friederich, P., \u0026amp; Brabec, C. J. (2024). Inverse design workflow discovers hole-transport materials tailored for perovskite solar cells. Science, 386(6727), 1256-1264(2024). https://doi.org/10.1126/science.ads0901\u003c/li\u003e\n \u003cli\u003eMicrosoft (2025). AI meets materials discovery: The vision behind MatterGen and MatterSim. https://www.microsoft.com/en-us/research/story/ai-meets-materials-discovery/ (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eZeni, C., Pinsler, R., Z\u0026uuml;gner, D., Fowler, A., Horton, M., Fu, X., Wang, Z., Shysheya, A., Crabb\u0026eacute;, J., Ueda, S., Sordillo, R., Sun, L., Smith, J., Nguyen, B., Schulz, H., Lewis, S., Huang, C.-W., Lu, Z., Zhou, Y., Yang, H., Hao, H., Li, J., Yang, C., Li, W., Tomioka, R., \u0026amp; Xie, T. (2025). A generative model for inorganic materials design. Nature 639, 624\u0026ndash;632. https://doi.org/10.1038/s41586-025-08628-5\u003c/li\u003e\n \u003cli\u003eYang, H., Hu, C., Zhou, Y., Liu, X., Shi, Y., Li, J., Li, G., Chen, Z., Chen, S., Zeni, C., Horton, M., Pinsler, R., Fowler, A., Z\u0026uuml;gner, D., Xie, T., Smith, J., Sun, L., Wang, Q., Kong, L., Liu, C., Hao, H., \u0026amp; Lu, Z. (2024). MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures. arXiv:2405.04967. https://doi.org/10.48550/arXiv.2405.04967\u003c/li\u003e\n \u003cli\u003ePham, G. K., Nguegang Tsopmo, U. E., Segura Cubells, M. (2025). Artificial Intelligence in Materials Science Education and Research \u0026amp; Development. The 2025 International Scientific Conference on Media Education in the age of AI and Digital Transformation, 25.12.2025, Hanoi, Vietnam. Conference proceedings, ISBN: 978-604-45-2379-8.\u003c/li\u003e\n \u003cli\u003eSwayne M. (2025). What is MatterGen? Microsoft\u0026rsquo;s generative AI could transform materials research. https://theaiinsider.tech/2025/01/17/what-is-mattergen-microsofts-generative-ai-could-transform-materials-research/ (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eXie, T. (2024). MatterGen: A Generative Model for Materials Design. https://www.microsoft.com/en-us/research/articles/mattergen-a-generative-model-for-materials-design/ (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eMicrosoft/MatterGen. (2024). MatterGen [Software]. GitHub. https://github.com/microsoft/mattergen (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eMicrosoft/MatterSim. (2024). MatterSim [Software]. GitHub. https://github.com/microsoft/mattersim (accessed on 24.10.2025)\u003c/li\u003e\n \u003cli\u003eZhang, L., Rao, A., Agrawala, M. (2023). Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF International Conference on Computer Vision. 3836\u0026ndash;3847. https://doi.org/10.1109/ICCV51070.2023.00355\u003c/li\u003e\n \u003cli\u003eMicrosoft/MatterSim. (2024). MatterSim Documentation: Getting Started. GitHub. https://microsoft.github.io/mattersim/user_guide/getting_started.html (accessed on 14.11.2025)\u003c/li\u003e\n \u003cli\u003eKittel, C. (2005). Introduction to solid state physics (8th ed.). Hoboken, NJ: Wiley.\u003c/li\u003e\n \u003cli\u003ePhonopy (2025). Workflow. https://phonopy.github.io/phonopy/workflow.html (accessed on 15.11.2025)\u003c/li\u003e\n \u003cli\u003eIbach, H., L\u0026uuml;th, H. (2009). Festk\u0026ouml;rperphysik: Einf\u0026uuml;hrung in die Grundlagen (7. Aufl.). Berlin: Springer.\u003c/li\u003e\n \u003cli\u003eDeepWiki (2025). Microsoft/mattersim: Phonon Calculations. https://deepwiki.com/microsoft/mattersim/6.2-phonon-calculations (accessed on 15.11.2025)\u003c/li\u003e\n \u003cli\u003eCASTEP Documentation (2023). Calculating phonon spectra for ferromagnetic iron. https://www.tcm.phy.cam.ac.uk/castep/documentation/WebHelp/content/tutorials/casteptutorial/castep_fdphonons_tut.htm (accessed on 15.11.2025)\u003c/li\u003e\n \u003cli\u003eSade, Y. A., Babaji, G., Lawal, M. A. (2017). Phonon dispersion relation and density of states in some selected FCC metal crystals. IOSR Journal of Applied Physics, 9(5), 47\u0026ndash;51. https://doi.org/10.9790/4861-0905024751\u003c/li\u003e\n \u003cli\u003eMondal, B., Mukherjee, T., Finch, N. W., Saha, A., Gao, M. Z., Palmer, T. A., DebRoy, T. (2023). Vapor pressure versus temperature relations of common elements. Materials, 16(50), 12345. https://doi.org/10.3390/ma16010050\u003c/li\u003e\n \u003cli\u003eL\u0026auml;pple, V., Drube, B., Wittke, G., Kammer, C. (2015). Werkstofftechnik Maschinenbau. Haan-Gruiten: Verlag Europa-Lehrmittel. ISBN 978-3-8085-5265-0.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"humancentric-intelligent-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Human-Centric Intelligent Systems](https://link.springer.com/journal/44230)","snPcode":"44230","submissionUrl":"https://submission.springernature.com/new-submission/44230/3","title":"Human-Centric Intelligent Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Artificial Intelligence, Machine Learning, MatterGen, MatterSim, Computational Materials Discovery","lastPublishedDoi":"10.21203/rs.3.rs-8143147/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8143147/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eArtificial Intelligence (AI) is reshaping materials science by accelerating discovery and enhancing education through data-driven approaches. This study explores the integration of Microsoft\u0026rsquo;s open-source tools, MatterGen and MatterSim, into academic and research workflows. MatterGen employs generative diffusion models to design novel inorganic materials, guided by property constraints and fine-tuned using large datasets. MatterSim, a machine-learning-based simulation engine, predicts thermodynamic, mechanical, and structural properties under realistic conditions with exceptional speed and accuracy.\u003c/p\u003e \u003cp\u003eThis work does not aim to propose a new ML model or benchmark, but to explore human-centric integration of MatterSim into materials education and early-stage research. In the first phase of this project, the focus was on evaluating MatterSim through simulations such as structural relaxation, phonon analysis, and molecular dynamics, supported by a custom-developed graphical user interface. The initial results showed good agreement between MatterSim predictions and reference data, confirming its reliability for educational and research purposes. Key benefits include a better understanding of material properties in a short time, improved accessibility for students and researchers to computational material discovery using AI, and reduced development time and costs.\u003c/p\u003e \u003cp\u003eHowever, limitations such as hardware requirements, incomplete thermodynamic workflows, and dependence on high-quality training data were identified. Future directions involve exploring the second Microsoft\u0026rsquo;s tool, MatterGen, and combining generative and predictive models for rapid screening. This work underscores the transformative potential of AI in materials science and provides practical recommendations for its adoption in both academic and industrial contexts. The custom-developed graphical user interface, integrated with MatterSim and implemented in Python, is published on GitHub as open source (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git\u003c/span\u003e\u003cspan address=\"https://github.com/ED0400/Mattersim-v1.0.0-1M_1st_Experience.git\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e","manuscriptTitle":"Accelerating Computational Material Discovery and Learning: First Experiences with MatterSim","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-04 05:48:12","doi":"10.21203/rs.3.rs-8143147/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-04-06T06:51:47+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"110410270488958225836334330046395644397","date":"2026-03-20T16:16:00+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-18T23:48:43+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-10T12:04:06+00:00","index":"","fulltext":""},{"type":"submitted","content":"Human-Centric Intelligent Systems","date":"2026-02-10T04:06:08+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"humancentric-intelligent-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [Human-Centric Intelligent Systems](https://link.springer.com/journal/44230)","snPcode":"44230","submissionUrl":"https://submission.springernature.com/new-submission/44230/3","title":"Human-Centric Intelligent Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Open","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e6e5cb79-60cf-44b5-9b70-82e84537e4a2","owner":[],"postedDate":"May 4th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-05-04T05:48:13+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-04 05:48:12","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8143147","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8143147","identity":"rs-8143147","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}