🔬 materials science

Kinetic Monte Carlo-Ising Machine Optimization for Atomistic Inverse Design of Solid Electrolytes

This paper presents a KMC-FMQA framework that combines kinetic Monte Carlo simulations with Ising-machine-based optimization to efficiently identify dopant configurations in solid electrolytes, successfully demonstrating an order-of-magnitude improvement in ionic conductivity for yttria-stabilized zirconia compared to random and experimental benchmarks.

Original authors: Ai Koizumi, Tomofumi Tada, Ryo Tamura

Published 2026-07-01

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ai Koizumi, Tomofumi Tada, Ryo Tamura

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to bake the perfect loaf of bread, but instead of flour and water, your ingredients are tiny atoms. Specifically, you are designing a "solid electrolyte," a material used in advanced fuel cells that acts like a highway for oxygen ions to travel through. The goal is to make this highway as fast as possible so electricity can be generated efficiently.

The problem is that the "traffic" (ionic conductivity) depends entirely on how you arrange the atoms. In a standard block of this material, the atoms are scattered randomly, like people milling about in a crowded room. This works okay, but it's not the fastest route. The researchers wanted to find the perfect arrangement of atoms to create a super-highway, but there are so many possible ways to arrange them that it's like trying to find a single specific grain of sand on all the beaches on Earth. You can't just check every possibility; it would take longer than the universe has existed.

The Solution: A Smart Team of Two

To solve this impossible puzzle, the authors created a team-up between two digital tools, acting like a Chef and a Taste-Tester.

The Chef (Kinetic Monte Carlo or KMC): This tool is the expert simulator. If you give it a specific arrangement of atoms, it can calculate exactly how fast the oxygen ions will move through it. However, the Chef is slow and can only taste one specific recipe at a time. It can't invent new recipes; it just evaluates the ones you hand it.
The Taste-Tester (FMQA or Ising Machine): This is a super-smart AI guesser. It looks at the results from the Chef and tries to learn the "flavor profile" of a fast highway. Instead of tasting every single possibility, it builds a mental map (a surrogate model) to predict which new arrangement of atoms will likely be the fastest. It then proposes a new recipe to the Chef.

How They Work Together

The process works like a loop:

The Taste-Tester suggests a few random atom arrangements.
The Chef simulates them and reports back: "This one is slow, but that one is fast!"
The Taste-Tester learns from this data. "Okay, it seems like having certain atoms clustered together helps."
The Taste-Tester uses this new knowledge to propose a better arrangement, hoping to beat the previous record.
The Chef tests this new one, and the cycle repeats.

The Secret Shortcut: Building with Blocks

Because the number of possible arrangements is still too huge, the researchers introduced a clever shortcut. Instead of moving individual atoms one by one, they grouped them into "blocks" (like Lego bricks). They realized that the material could be built using just three types of these blocks. This turned a chaotic mess of millions of possibilities into a manageable set of 81 distinct "block recipes."

They then split the work, running 81 different teams of the Chef and Taste-Tester in parallel, each trying to find the best version of their specific block recipe.

The Result

After running this loop, the team found a specific arrangement of atoms that was ten times faster at conducting electricity than the standard random arrangement and ten times faster than what has been measured in real-world experiments so far.

They discovered that the "perfect" arrangement isn't random at all. Instead, the atoms form neat, layered patterns near the edges of the material, creating a smooth, unobstructed path for the ions to zoom through.

Why This Matters (According to the Paper)

This study proves that you can use this "Chef and Taste-Tester" loop to reverse-engineer materials. Instead of just guessing what a material looks like and seeing how it works, you can start with the goal (maximum speed) and let the computer work backward to tell you exactly how the atoms should be arranged to achieve it. This opens the door to designing better fuel cells by finding the microscopic structures that nature hasn't quite stumbled upon yet.

Technical Summary: Kinetic Monte Carlo–Ising Machine Optimization for Atomistic Inverse Design of Solid Electrolytes

Problem Statement
The performance of solid oxide fuel cells (SOFCs) relies heavily on the ionic conductivity of their solid electrolytes. While the dopant concentration in materials like 8 mol % yttria-stabilized zirconia (8YSZ) is well-studied, the specific microscopic arrangement of dopants (configuration) significantly influences conductivity, particularly in miniaturized devices or thin films. Previous work established that ionic conductivity depends systematically on dopant configuration, yet experimentally elucidating this structure–conductivity relationship remains challenging. The core problem addressed is "atomistic inverse design": identifying the specific dopant configuration that maximizes ionic conductivity. This is a massive combinatorial optimization problem where the configuration space is too vast for exhaustive search, and conventional black-box optimization (BBO) methods like Bayesian optimization become intractable for large multicomponent systems.

Methodology: The KMC–FMQA Framework
The authors propose a hybrid framework combining Kinetic Monte Carlo (KMC) simulations with Factorization Machine with Quadratic-optimization Annealing (FMQA), an Ising-machine-based BBO algorithm. The workflow operates as follows:

Forward Evaluation (KMC): KMC simulations calculate the ionic conductivity (drift current) for a specific dopant configuration. The simulations focus on oxygen-ion migration through six-site paths, where migration barriers are determined by the local dopant environment.
Surrogate Modeling (FM): A Factorization Machine (FM) acts as a surrogate model, learning the relationship between dopant configurations (encoded as bit strings) and their resulting conductivities from a small dataset.
Optimization (Ising Machine): The trained FM is used by an Ising machine to search for new configurations that maximize the predicted conductivity. This step solves a Quadratic Unconstrained Binary Optimization (QUBO) problem subject to physical constraints.
Iterative Loop: The conductivity of the newly proposed configuration is calculated via KMC and added to the training dataset, refining the FM for the next iteration.

Key Technical Innovations
To manage the combinatorial explosion in a 16-layer bulk 8YSZ supercell, the authors introduced two critical strategies:

Block Model Representation: Instead of optimizing individual atomic sites, the system is constructed from "blocks" of four lattice sites. Three block types are defined based on dopant content: $Zr_4$ , $Y_2Zr_2$ , and $Y_4$ . This reduces the search space and maps the problem to a QUBO format suitable for Ising machines.
Composition-Based Partitioning: To maintain the 8YSZ stoichiometry, the authors enumerated 81 possible block compositions (varying the ratio of block types). The vast configuration space was partitioned into 81 independent subspaces, each optimized in parallel. This approach proved more effective than simply increasing the number of trials in a single, unpartitioned search space.

Results
The framework was applied to optimize dopant configurations in a 16-layer region of a bulk 8YSZ supercell.

Performance: The KMC–FMQA framework successfully identified configurations with ionic conductivities significantly higher than random distributions. The best configuration achieved a conductivity of 0.423 S cm⁻¹.
Comparison: This result represents an order-of-magnitude improvement over the average conductivity of random dopant distributions (0.06 S cm⁻¹) and the experimentally reported conductivity of bulk 8YSZ (0.0505 S cm⁻¹).
Structural Insights: The high-conductivity configurations exhibited a specific structural feature: dopants were highly concentrated near the edges of the 16-layer region (YSZ/ZrO₂ interfaces). The authors suggest that these layered configurations promote oxygen vacancy accumulation along the layers, modifying electrostatic contributions to migration barriers and facilitating ionic transport.
Benchmarking: Comparative tests against an all-atom model and a block model without partitioning confirmed that the block-based representation combined with composition-based partitioning yielded the best optimization performance.

Significance and Claims
The paper presents this work as a "proof of concept" demonstrating that the KMC–FMQA framework enables efficient atomistic inverse design for solid electrolytes. The authors claim that this approach allows for the rapid identification of high-conductivity configurations that would be difficult to find through random sampling or conventional optimization.

The authors state that, when combined with experiments, this framework has the potential to enable the inverse estimation of microscopic structures from measured conductivity values, thereby providing deeper insight into ionic transport mechanisms. They further note that the approach is extensible to electrolyte-electrode interfaces, offering a pathway for broader atomistic materials design to optimize SOFC performance. The study does not claim to have experimentally synthesized these specific configurations but focuses on the computational capability to identify them.

More like this