Materials Informatics Advanced Practical: Course Overview#
Welcome to this small practical course on computational materials design! In this course, you’ll learn how to use ml/ai and informatics tools to learn about the materials design process with relevant examples enhancing your understanding behind how batteries, solar cells, catalysts, semiconductors and a variety of other materials applications can be designed virtually before being synthesised and utilised in the real world.
At the heart of this journey is SMACT (Semiconducting Materials from Analogy and Chemical Theory), a powerful Python toolkit that enables rapid screening of millions of potential materials using chemical rules and data-driven approaches. You’ll also explore cutting-edge AI tools like Chemeleon, which can generate crystal structures from simple text descriptions. The course culminates with comprehensive tutorials on advanced computational methods including MACE (machine learning force fields) and Density Functional Theory, providing you with a complete toolkit for modern materials research.
Course Overview#
This course is structured to provide both theoretical knowledge and practical skills in materials informatics. Below is an overview of the key topics we will cover:
1. Combinatorial Explosion and Chemical Space Generation#
Description:
Understand the concept of combinatorial explosion in materials science and its implications for materials discovery.
Utilise SMACT’s capabilities to generate extensive lists of elemental compositions.
Prepare these compositions as inputs for machine learning models and other screening workflows.
Learning Outcomes:
Grasp the challenges posed by vast chemical spaces.
Efficiently generate and manage large datasets of potential material compositions.
2. Application of Chemical Filters#
Description:
Learn about chemical filters—a set of rules applied to reduce chemical spaces to viable candidates.
Use a modified
smact_filterfunction to apply these rules to generated compositions.Convert filtered results into dataframes with features suitable for machine learning algorithms.
Learning Outcomes:
Apply chemical filters to streamline candidate selection.
Prepare datasets for machine learning applications.
3. Compositional Screening and Phase Diagram Analysis#
Description:
Use SMACT to screen chemical spaces for specific element combinations (e.g., {Cu, Ti, O}).
Generate phase diagrams illustrating all allowable and forbidden composition possibilities.
Interpret these diagrams to identify promising material candidates.
Learning Outcomes:
Perform targeted compositional screenings.
Analyse and interpret phase diagrams for material selection.
4. Stoichiometry Screening#
Description:
Explore methods to find compounds that satisfy specific stoichiometries (e.g., perovskite ABX₃).
Utilise a modified SMACT filter function to identify such compounds.
Assess the potential of these compounds for further study.
Learning Outcomes:
Conduct stoichiometry-based screenings.
Identify and evaluate compounds matching desired stoichiometric formulas.
5. Composition to Structure Techniques#
5.1 Structure Prediction with SMACT#
Description:
Learn about SMACT’s structure prediction submodule based on ionic substitution, inspired by Hautier et al.’s 2011 paper.
Set up a database of SMACT-compatible structures.
Assign compositions to known crystal structures using the structure prediction tool.
Learning Outcomes:
Utilise ionic substitution methods for structure prediction.
Build and manage structural databases for materials exploration.
5.2 Crystal Structure Generation with Chemeleon#
Description:
Explore Chemeleon, a modern generative AI model that creates crystal structures from text descriptions.
Learn how Chemeleon differs from traditional substitution-based methods by generating structures from scratch.
Practice both Crystal Structure Prediction (CSP) for specific formulas and De Novo Generation (DNG) for exploring unknown chemical space.
Compare the complementary strengths of SMACT’s physics-based approach with Chemeleon’s AI-driven generation.
Learning Outcomes:
Generate crystal structures using natural language prompts and chemical formulas.
Understand the shift from substitution to generation in materials discovery.
Evaluate when to use traditional vs. AI-driven structure prediction methods.
6. Advanced Methods#
Description:
Dive into advanced topics through comprehensive documentation and tutorials.
Density Functional Theory (DFT): Understand the theoretical foundations of quantum mechanical calculations with detailed guides on DFT theory and VASP capabilities.
Machine Learning Force Fields: Explore MACE (Multi-Atomic Cluster Expansion) through practical tutorials and theoretical explanations, learning how to achieve DFT-level accuracy at classical simulation speeds.
Thermodynamics: Understand the role of thermodynamic principles in materials discovery.
Synthesisability: Coming soon :) .
Learning Outcomes:
Apply advanced computational methods to materials science problems.
Understand when to use DFT (VASP) versus machine learning approaches (MACE).
Set up and run MACE simulations for rapid materials screening.
Integrate thermodynamic principles with machine learning approaches.
Course Objectives#
By the end of this course, you will be able to:
Generate and Analyse Chemical Spaces: Efficiently create large chemical composition datasets and analyse them using SMACT.
Apply Chemical Filters: Implement chemical filtering techniques to narrow down potential material candidates.
Perform Compositional and Stoichiometry Screening: Conduct targeted screenings to identify promising materials based on composition and stoichiometry.
Utilise Structure Prediction Tools: Predict and generate crystal structures using SMACT and Chemeleon.
Understand Advanced Computational Methods: Grasp the basics of thermodynamics, Machine Learning Force Fields, DFT, and synthesisability in the context of materials science.
Integrate Machine Learning Techniques: Apply machine learning algorithms to materials informatics workflows.
Course Format#
This is a self-paced practical course built around:
Interactive Jupyter Notebooks: Each topic includes a hands-on notebook with code examples and exercises.
Progressive Learning Path: Starting from basic chemical screening to advanced AI-driven structure generation.
Real-World Applications: Examples drawn from battery materials, solar cells, and other energy applications.
Immediate Practice: Run code locally or in Google Colab to see results in real-time.
Prerequisites#
Basic Chemistry Knowledge: Understanding of chemical formulas, oxidation states, and crystal structures.
Python Fundamentals: Ability to run Python scripts, work with lists/dictionaries, and use basic libraries like NumPy.
No ML Experience Required: We’ll introduce machine learning concepts as needed.
Curiosity: Enthusiasm for learning and applying the skills you learn here in your future materials design workflows.
Materials and Resources#
SMACT Toolkit: github.com/WMD-group/SMACT - Core screening toolkit
Chemeleon: github.com/hspark1212/chemeleon-dng - AI structure generation
Materials Project API: Access to experimental crystal structure data
Course Repository: All notebooks, data files, and setup scripts
Key Papers: Referenced throughout with DOI links for deeper understanding
How to Use This Course#
Start with Setup: Follow the setup instructions to install all required tools.
Work Through Topics Sequentially: Each topic builds on the previous ones.
Run All Code Examples: Understanding comes from doing - run and modify the examples.
Complete the Exercises: Each notebook includes exercises to reinforce learning.
Explore Your Own Ideas: Use the tools to investigate materials relevant to your interests.
If you have any questions, don’t hesitate to reach out and if there are any urgent issues with the notebooks drop Ry an email - napo.nduma22@imperial.ac.uk