Chapter 1: Why Materials Informatics Now?

This chapter organizes the history of materials development and the limitations of conventional methods to build an intuitive understanding of why MI is needed today. We will survey the trends and success stories since the Materials Genome Initiative (MGI) and create a roadmap for subsequent learning.

💡 Note: MI is "a data-driven approach to expand exploration using data + machine learning." It moves beyond reliance on chance and experience toward reproducible decision-making.

Learning Objectives

By reading this chapter, you will be able to: - Understand the historical evolution of materials development (from the Bronze Age to the present) - Explain the limitations and challenges of conventional materials development - Understand the social and technological background for the need for MI - Learn about the difficulties and possibilities of materials development through specific examples of lithium-ion battery development

1.1 The History of Materials Development: 5000 Years of Trial and Error

Human civilization has always evolved alongside materials. The types and performance of materials we use have characterized each era.

Ancient Times: The Era of Accidental Discovery

The Bronze Age (around 3000 BCE)

Bronze (an alloy of copper and tin), humanity's first alloy, was likely a fortuitous product. Copper and tin ores were smelted together in a mixed state, and by chance, a metal much harder than pure copper was obtained. This discovery enabled humanity to transition from the Stone Age to the Metal Age.

The development method was entirely accidental. The development period spanned several hundred years to discover the optimal composition ratio (approximately 90% copper, 10% tin). Knowledge accumulation occurred only through oral tradition and empirical rules.

The Iron Age (around 1200 BCE)

The establishment of iron smelting technology made harder and more abundant materials available. However, establishing technologies to control iron properties (quenching and tempering) required several more centuries.

Modern Era: Development Based on Empirical Rules (1800-1950s)

The Age of Steel (1800s)

During the Industrial Revolution, Henry Bessemer invented the Bessemer process (1856), enabling mass production of steel. However, this invention was fundamentally also a product of trial and error.

The development method relied on trial and error through experimentation. The development period required 5-10 years for one new material. Knowledge accumulation was based on empirical rules and observational records.

Discovery of Stainless Steel (1913)

British metallurgist Harry Brearley was researching iron-chromium alloys to improve the corrosion resistance of gun barrels. During experiments, he accidentally discovered steel that did not rust (stainless steel). This discovery was also essentially based on chance and empirical rules.

Contemporary Era: The Beginning of Theory-Based Design (1950s-Present)

Silicon Semiconductors (1950s)

With the establishment of transistors (invented in 1947) and silicon semiconductor technology, the foundation for the information age was built. Around this time, materials design based on theoretical foundations such as quantum mechanics began.

The development method evolved to theory-based design combined with experimental validation. The development period extended to 10-20 years for one new material. Knowledge accumulation shifted to scientific papers, patents, and databases.

Polymer Materials and Composite Materials (1990s onward)

Composite materials such as carbon fiber reinforced plastics (CFRP) began to be adopted in aircraft and automobiles. These materials are designed using theoretical calculations and simulations. However, ultimately, experimental validation remains essential.

Challenges Visible from History

Looking back on 5000 years of materials development history reveals the following challenges:

1. Dependence on Chance: Many important discoveries were fortuitous products
2. Time-Consuming Process: Several years to decades for one material development
3. Limited Search Range: Only exploring what researchers can conceive
4. Dispersed Knowledge: Knowledge is scattered among individuals and organizations, making systematic accumulation difficult

Question: What if we could systematically search for materials using computers?

This is the starting point of the next-generation materials development method, Materials Informatics (MI).

1.2 Limitations of Conventional Materials Development

Modern materials development has become far more advanced than in ancient times. However, significant challenges remain.

Challenge 1: Time-Consuming

Typical Materials Development Timeline

Years 1-2: Literature survey and theoretical investigation
  ↓
Years 3-5: Synthesis and evaluation of candidate materials (10-50 types)
  ↓
Years 6-10: Optimization and property characterization
  ↓
Years 11-15: Establishment of mass synthesis processes for practical use
  ↓
Years 16-20: Practical application and commercialization

Result: Practical application of new materials takes an average of 15-20 years[1,2].

Challenge 2: High Cost

Cost to Evaluate One Material

Material synthesis costs 100,000 to 1,000,000 yen for reagents and equipment usage fees. Property evaluation requires 500,000 to 5,000,000 yen for measurement equipment usage fees and analysis costs. Personnel costs for 1 researcher working 1-2 weeks amount to 500,000 to 1,000,000 yen.

Total: 1 million to 7 million yen per material

A research laboratory with an annual budget of 30 million yen can only evaluate 10-30 materials per year.

Challenge 3: Limited Search Range

Material Combinations are Astronomical Numbers

The periodic table contains about 90 practical elements. Binary alloy combinations number about 4,000 types, while ternary alloy combinations reach about 117,000 types. Quaternary alloy combinations expand to about 2.9 million types.

Furthermore, considering variations in composition ratios and crystal structures, there are essentially infinite combinations.

With conventional methods, researchers select tens to hundreds of materials based on experience and intuition for experimentation. In other words, only a tiny fraction of the vast possibilities can be explored.

Challenge 4: Dependence on Experience and Intuition

Tacit Knowledge of Expert Researchers

Experts rely on knowledge such as "This element combination tends to be unstable," "This crystal structure is favorable for ion conduction," and "Sintering at this temperature produces good properties."

Such tacit knowledge is extremely valuable, but has the following problems:

1. Difficult to Systematize: Based on personal experience, making it hard to share
2. Reproducibility Issues: Results can differ among researchers even under the same conditions
3. Time for Training Juniors: Expertise requires over 10 years of experience
4. Presence of Bias: Bound by existing knowledge, potentially missing innovative discoveries

1.3 Case Study: 20 Years of Lithium-Ion Battery Development

The lithium-ion battery development story is a typical example demonstrating the difficulties of conventional materials development and the importance of persistent research. Let's examine this technology, which won the 2019 Nobel Prize in Chemistry, in detail.

Phase 1: Basic Research (1970s)

Background: Energy Crisis

The oil shocks of the 1970s increased interest in energy storage technologies that did not depend on petroleum.

Dr. John Goodenough's Challenge

Dr. Goodenough at Oxford University believed that lithium-containing oxide materials were promising for energy storage.

Material Candidates Explored: Over 100 types

• LiMO₂ (M = Ti, V, Cr, Mn, Fe, Co, Ni)
• LiM₂O₄ (M = Mn, Co)
• Various crystal structures (layered, spinel, olivine)

Discovery (1980): LiCoO₂ (lithium cobalt oxide)

• Features: Layered structure, Li ions can move between layers
• Theoretical capacity: 274 mAh/g
• Operating voltage: About 4V (high voltage for that time)

However, many challenges remained to commercialize this material.

Phase 2: Anode Material Development (1980s)

Challenge: Danger of Metallic Lithium

Initially, metallic lithium was used for the anode, but had the following problems:

1. Dendrite Formation: With repeated charging and discharging, needle-shaped lithium crystals (dendrites) grow
2. Short Circuit Risk: When dendrites reach the cathode, short circuits occur with fire danger
3. Cycle Life: Degradation after several dozen charge-discharge cycles

Dr. Akira Yoshino's Solution (1985)

Dr. Yoshino at Asahi Kasei conceived the idea of using carbon material (graphite) as a lithium ion intercalation material.

Material Candidates Explored: Over 50 types

• Various graphite materials
• Amorphous carbon
• Graphite intercalation compounds

Achievements: - Graphite safely intercalates Li ions between layers - No dendrite formation - Cycle life improved to several hundred cycles

Phase 3: Electrolyte Optimization (Late 1980s)

Challenge: Electrolyte Stability

An electrolyte that operates stably between the cathode (4V) and anode (0V vs Li/Li⁺) was needed.

Material Candidates Explored: Over 100 types

• Various organic solvent combinations
• Lithium salts (LiPF₆, LiBF₄, LiClO₄, etc.)
• Additives (controlling SEI film formation)

Discovery of Optimal Solution: - Ethylene carbonate (EC) + Diethyl carbonate (DEC) - Lithium salt: LiPF₆ - This combination achieved stable charging and discharging

Phase 4: Practical Application (1991)

Commercialization by Sony

In 1991, Sony commercialized the world's first lithium-ion battery as a video camera battery.

Specifications (1991 First Generation): - Energy density: About 200 Wh/kg (about twice that of nickel-metal hydride batteries) - Cycle life: Over 500 cycles - Operating voltage: 3.7V

Time and Cost of Development

Time: About 20 years from the start of basic research (1970s) to commercialization (1991)

Researchers: Led by Drs. Goodenough, Whittingham, and Yoshino, hundreds of researchers worldwide were involved

Total Materials Explored: Estimated over 500 types

Failed Experiments: Over thousands of times

Question: What if MI had existed?

If modern MI technology had existed in the 1970s:

1. Narrowing Material Candidates: Machine learning predicts 100 promising materials in a few days
2. Electrolyte Optimization: Bayesian optimization discovers optimal composition with about 20 experiments
3. Development Period: Could be shortened to an estimated 5-7 years (less than 1/3)

This is not science fiction, but actually possible with modern MI technology.

1.4 Conventional Methods vs MI: Workflow Comparison

As seen in the lithium-ion battery example, conventional materials development is time-consuming and costly. Here, let's visually compare the workflows of conventional and MI methods.

Workflow Comparison Diagram

Quantitative Comparison

Timeline Comparison Example

Evaluating 100 materials with conventional methods: - 1 material × 4 weeks = 100 materials × 4 weeks = 400 weeks = about 8 years

Evaluating 100 materials with MI methods: - Data collection and model building: 2 weeks - Predict 10,000 types: 1 day - Experiment on top 100 types: 100 materials × 2 weeks = 200 weeks = about 4 years - However, with parallel experiments and robotic automation: 6 months-1 year

Time Reduction: 8 years → 6 months-1 year = 87-93% reduction

1.5 Column: A Day in the Life of a Materials Scientist

Let's look at specific stories of how the materials development field has changed.

1985: The Era of Conventional Methods

Professor Tanaka's (45 years old) Day

9:00 - Arrive at Laboratory Yesterday's sample synthesis completed. Remove from furnace and begin cooling.

10:00 - Sample Property Evaluation Go to experimental room for X-ray diffraction measurement. Measurement takes 3 hours. Read papers in the meantime.

14:00 - Data Analysis Manually analyze X-ray diffraction pattern. 2 hours to identify crystal structure.

16:00 - Plan Next Experiment Based on today's results, consider the next sample composition. Based on experience, decide to try material with slightly changed composition.

17:00 - Prepare Sample Prepare new sample for tomorrow. Weigh reagents, mix, set in furnace.

18:00 - Record in Experimental Notebook Record today's results in detail in handwritten experimental notebook.

19:00 - Leave Work

Day's Achievement: Evaluated 1 type of material, prepared next 1 type

Month's Achievement (20 days): About 20 types of materials evaluated

Year's Achievement: About 200 types of materials evaluated (actually about 150 types due to equipment troubles and vacations)

2025: The MI Era

Associate Professor Sato's (38 years old) Day

9:00 - Arrive at Laboratory First, check results of 10 sample types executed by automated experimental equipment overnight. Data is automatically saved in cloud database.

9:30 - AI Data Analysis Machine learning model automatically identifies crystal structures and predicts properties. Analysis of 10 types of data completed in 10 minutes.

10:00 - Predict Next Experiment Candidates Bayesian optimization algorithm proposes 20 promising types to try next from a database of 100,000 material types. Prediction takes 5 minutes.

10:30 - Review Top Candidates Examine the proposed 20 types with human eyes. Leveraging materials science knowledge, select particularly promising 10 types.

11:00 - Set Experimental Conditions Input synthesis conditions for the selected 10 types into automated experimental equipment.

11:30 - Research Meeting Discuss this week's progress with students. Examine validity of AI prediction results and consider next research directions.

13:00 - Paper Writing Increased time for paper writing by utilizing overnight experiment time.

15:00 - Automated Equipment Maintenance Check equipment operation and replace consumables.

16:00 - Train New Model Add new data obtained this week and retrain machine learning model. Prediction accuracy further improves.

17:00 - Leave Work

Day's Achievement: Evaluated 10 types of materials, set next 10 types for automated experiments

Month's Achievement (20 days): About 200 types of materials evaluated

Year's Achievement: About 2,000 types of materials evaluated (operates on weekends and holidays due to automation)

Key Points of Change

Important Point: MI is a tool to support, not replace researchers. Both Professor Tanaka's experience and Associate Professor Sato's judgment are indispensable, but Associate Professor Sato can explore more materials, more efficiently with AI support.

1.6 Why "Now" MI?: Three Tailwinds

The concept of MI itself has existed since the 1990s, but it was fully put into practical use only from the 2010s onward. Why "now"? There are three major factors.

Tailwind 1: Dramatic Improvement in Computing Performance

Benefits of Moore's Law

• 1990: First-principles calculation of one material takes several weeks
• 2000: Calculation of one material takes several days
• 2010: Calculation of one material takes several hours
• 2020: Calculation of one material takes several minutes to tens of minutes

Spread of Cloud Computing

• Cloud services like AWS and Google Cloud make high-performance computing accessible to everyone
• Use supercomputer-level computing resources for tens to hundreds of yen per hour
• Parallel computing enables prediction of 10,000 material types in one day

Utilization of GPU (Graphics Processing Units)

• GPU computing became mainstream with the spread of deep learning
• Can train machine learning models 100+ times faster than CPUs
• GPU manufacturers like NVIDIA provide research GPUs

Tailwind 2: Enrichment of Materials Databases

Materials Project (started 2011)

• Operated by Lawrence Berkeley National Laboratory
• Materials database using first-principles calculations
• Over 140,000 types of material data (as of 2024)[3]
• Diverse properties including crystal structure, energy, band gap, elastic constants
• Free access (API also provided)

Other Major Databases

Open Science Trends

• Standardization of research data publication
• Culture of publishing datasets along with papers
• Anyone can access data on platforms like GitHub and Zenodo

Tailwind 3: Increased Social Urgency

US Materials Genome Initiative (MGI) (2011)

A national project started by the Obama administration. With the goal of halving materials development time, research investment was accelerated in both public and private sectors.

Goals: - Materials development period: 20 years → 10 years or less - Integration of computation, experiment, and data - Annual budget: About $100 million (about 10 billion yen)

Response to Climate Change

• 2015 Paris Agreement: Limit global warming to within 2°C
• Urgent development of renewable energy, energy storage, and CO₂ reduction materials
• Improve lithium-ion battery performance (extend electric vehicle range)
• Improve solar cell efficiency (reduce power generation costs)

Spread of Electric Vehicles (EVs)

• 2020s: EV adoption accelerating worldwide
• China, EU, and US planning gasoline vehicle sales regulations
• Need for development of higher-performance battery materials
• Conventional methods cannot keep up with demand

Global Competition

• China: Massive investment in materials research as national strategy
• Europe: Supporting materials research through Horizon Europe program
• Japan: Cabinet Office "Material Innovation Enhancement Strategy" (2021)

Conclusion: MI is a technology needed precisely now, when technological maturity and social necessity are simultaneously satisfied.

1.7 MI Standard Pipeline: Overall Picture

So far, we have examined the necessity and potential of MI. So how are MI projects actually conducted? Let's look at the standard pipeline.

MI Standard Workflow

Details of Each Step

Step 0: Problem Formulation - Purpose: Clearly define the problem to solve - Content: Set target properties, constraints, and success criteria (KPI; Key Performance Indicator) - Time: 1-2 weeks - Example: "Search for materials with band gap of 2.0-2.5 eV"

Step 1: Data Collection - Purpose: Collect training data for machine learning - Sources: Materials Project, OQMD, literature, experimental data - Time: 1-4 weeks - Tools: pymatgen, API, Web scraping

Step 2: Feature Engineering - Purpose: Convert materials to numerical vectors - Content: Calculate composition descriptors, structure descriptors - Time: Several days-1 week - Tools: Matminer, Element Property

Step 3: Model Building - Purpose: Train machine learning model to predict properties - Methods: Random Forest, Neural Networks, Gradient Boosting - Time: Several hours-several days - Tools: scikit-learn, PyTorch, TensorFlow

Step 4: Prediction & Evaluation - Purpose: Evaluate model performance and predict new materials - Metrics: MAE (Mean Absolute Error), R² (coefficient of determination), RMSE (Root Mean Square Error) - Time: Several minutes-several hours - Criterion: R² > 0.7 (practical level)

Step 5: Experimental Validation - Purpose: Validate prediction results experimentally - Content: Synthesis, property measurement, comparison with predictions - Time: Several weeks-several months - Outcome: Add new experimental data to database

Iterative Cycle - Add experimental data to training data and retrain model - Prediction accuracy improves and search efficiency increases - Usually achieve goals in 3-5 cycles

Comparison with Conventional Methods

Important Points

1. Step 0 is Most Important: If problem formulation is inappropriate, all subsequent steps become wasted
2. Data Quality Determines Results: Garbage in, garbage out
3. Iteration is Key: Cannot create perfect model in one attempt. Repetition of experiment→learning is necessary
4. Human Judgment is Essential: AI proposes, but final judgment is made by materials scientists

We will learn this pipeline in detail from Chapter 2 onward.

1.8 Chapter Summary

What We Learned

1. History of Materials Development - From the Bronze Age to present, materials have supported civilization's development - Evolved from ancient chance to modern trial and error to contemporary theory-based design - However, development still takes 10-20 years

2. Limitations of Conventional Methods - Time: 4-8 weeks per material, 15-20 years to practical application - Cost: 1-7 million yen per material - Search Range: Only 10-100 types per year (just a fraction of possibilities) - Experience Dependence: Dependent on tacit knowledge of expert researchers

3. Lessons from Lithium-Ion Battery - 20 years from basic research (1970s) to commercialization (1991) - Over 500 types of materials through trial and error - Thousands of failed experiments - MI could reduce development period by 1/3

4. Advantages of MI - Annual exploration: 10-30 types → 100-2000 types (10-100x) - Development period: 15-20 years → 3-7 years (60-80% reduction) - Cost reduction: 90-99% reduction (utilizing computational predictions)

5. Why MI is Needed "Now" - Improved computing performance (Moore's Law, GPU, cloud) - Enriched materials databases (Materials Project, etc., over 140,000 types) - Social urgency (climate change, EV adoption, international competition)

Important Points

MI is a tool to support, not replace researchers. The combination of computational prediction and experimental validation is important. Data quality and quantity determine prediction accuracy, and both materials science and data science knowledge are necessary.

To the Next Chapter

Chapter 2 will examine the basic workflow of MI in detail, covering data collection methods, building machine learning models, prediction and screening, and experimental validation and data cycle.

We will also perform simple material prediction practice using Python.

Practice Problems

Problem 1 (Difficulty: easy)

Explain how development methods evolved through three eras in materials development history: the Bronze Age, Iron Age, and modern times.

Problem 2 (Difficulty: easy)

Calculate how long it takes to evaluate 100 types of materials using conventional materials development methods. Assume 4 weeks per material.

Problem 3 (Difficulty: medium)

In the development of lithium-ion batteries, explain specifically how the development process would have changed if MI technology had existed in the 1970s.

References

1. Ramprasad, R., Batra, R., Pilania, G., Mannodi-Kanakkithodi, A., & Kim, C. (2017). "Machine learning in materials informatics: recent applications and prospects." npj Computational Materials, 3(1), 54. DOI: 10.1038/s41524-017-0056-5

2. Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O., & Walsh, A. (2018). "Machine learning for molecular and materials science." Nature, 559(7715), 547-555. DOI: 10.1038/s41586-018-0337-2

3. Jain, A., Ong, S. P., Hautier, G., Chen, W., Richards, W. D., et al. (2013). "Commentary: The Materials Project: A materials genome approach to accelerating materials innovation." APL Materials, 1(1), 011002. DOI: 10.1063/1.4812323 Materials Project: https://materialsproject.org

4. Goodenough, J. B., & Park, K. S. (2013). "The Li-ion rechargeable battery: a perspective." Journal of the American Chemical Society, 135(4), 1167-1176. DOI: 10.1021/ja3091438

5. National Science and Technology Council (2011). "Materials Genome Initiative for Global Competitiveness." Executive Office of the President, USA. URL: https://www.mgi.gov/

Author Information

Author: MI Knowledge Hub Content Team Date Created: 2025-10-16 Version: 3.0 (Chapter 1 standalone version) Template: content_agent_prompts.py v1.0

Revision History: - 2025-10-16: v3.0 Chapter 1 standalone version created - Expanded from v2.1 Section 1 (58 lines) to 3,000-4,000 words - Added materials development history section (Bronze Age-present) - Detailed Li-ion battery case study (20-year development process) - Added workflow comparison Mermaid diagram - Added "A Day in the Life of a Materials Scientist" column (1985 vs 2025) - Added "Why MI Now" section (three tailwinds) - Added 3 practice problems

License: Creative Commons BY-NC-SA 4.0