High-Throughput Computation of Li-based Battery Material Databases: Chemistry-Processing-Property Relationships

Scott R. Broderick,Kaito Miyamoto,Krishna Rajan

DOI: https://doi.org/10.48550/arXiv.2203.08699

2022-03-16

Abstract:In this paper, we incorporate processing into the data driven property modeling of Li-based spinel battery materials. When considering spinel compounds of the form LiMe2O4 with Me as a metal or metals, there are 125,000 possible combinations assuming a maximum of three metallic elements. In this work, we focus on capacity for our predicted property, and increase the number of possible combinations to two million by incorporating processing into the modeling. Due to the non-linear relationship between processing and property, as well as ensuring proper sensitivity, a new approach which tracks the change with changing processing is introduced. This work provides an invaluable tool for guiding the next generation of battery experiments, providing a significant narrowing of the infinite design space.

Materials Science

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to incorporate processing parameters (such as annealing time and temperature) into high - throughput data - driven property modeling of lithium - based battery materials to predict the capacity of these materials. Specifically, the author focuses on lithium - based materials with a spinel structure (formula: LiMe₂O₄) and hopes to expand the possible combination space by introducing processing parameters, thereby predicting material properties more accurately.

### Main problems
1. **Challenges of introducing processing parameters**:
- Traditional material design methods mainly focus on chemical composition, while the influence of processing conditions on material properties is often ignored.
- The relationship between processing parameters and material properties is nonlinear, and the sensitivity of the model needs to be ensured.

2. **Exploration of large - scale combination space**:
- In the spinel compound LiMe₂O₄, the metal element Me can have multiple combinations, resulting in a very large number of possible combinations (125,000 assuming a maximum of three metal elements).
- By introducing processing parameters, the number of possible combinations further increases to 2 million.

3. **Improving prediction accuracy**:
- In order to accurately predict the capacity of materials, the author proposes a new method to capture the complex relationship between chemical composition and processing conditions through nonlinear manifold learning (such as the IsoMap algorithm).

### Solutions
- **Introducing processing parameters**: The author takes annealing time and temperature as additional controllable descriptors or parameters and constructs a matrix of "chemical descriptor set + processing".
- **Nonlinear manifold learning**: Use the IsoMap algorithm to perform dimensionality reduction on high - dimensional data and generate parameters that can capture nonlinear relationships.
- **Regression analysis**: Based on the generated parameters, perform regression analysis to establish a high - throughput model for predicting the capacity of materials.

### Significance
- **Guiding experimental design**: Through this method, researchers can quickly screen out the most promising materials and processing conditions, thereby guiding future experimental design.
- **Expanding the material database**: This method can significantly expand the knowledge base of battery materials and predict the properties of millions of materials that have not been experimentally measured.

### Formula example
In the IsoMap algorithm mentioned in the article, the parameterization process can be represented by the following formula:

\[
x_i \rightarrow y_i | y_i \in M_d, d < n, s.t. \forall i, j: \| x_i - x_j \|_\beta = \| y_i - y_j \|_\beta
\]

where:
- \( x_i \) and \( x_j \) are data points in high - dimensional space.
- \( y_i \) and \( y_j \) are corresponding points on the low - dimensional manifold.
- \( \beta \) is a norm, representing the pairwise geodesic distance \( d_{ij} \).

Through this method, the author has successfully solved the problem of incorporating processing parameters into the high - throughput prediction model of lithium - based battery materials.