Li$_x$CoO$_2$ phase stability studied by machine learning-enabled scale bridging between electronic structure, statistical mechanics and phase field theories

Gregory H. Teichert, Sambit Das, Muratahan Aykol, Chirranjeevi Balaji Gopal, Vikram Gavini, Krishna Garikipati

Abstract: Li$_xTM$O$_2$ (TM={Ni, Co, Mn}) are promising cathodes for Li-ion batteries, whose electrochemical cycling performance is strongly governed by crystal structure and phase stability as a function of Li content at the atomistic scale. Here, we use Li$_x$CoO$_2$ (LCO) as a model system to benchmark a scale-bridging framework that combines density functional theory (DFT) calculations at the atomistic scale with phase field modeling at the continuum scale to understand the impact of phase stability on microstructure evolution. This scale bridging is accomplished by incorporating traditional statistical mechanics methods with integrable deep neural networks, which allows formation energies for specific atomic configurations to be coarse-grained and incorporated in a neural network description of the free energy of the material. The resulting realistic free energy functions enable atomistically informed phase-field simulations. These computational results allow us to make connections to experimental work on LCO cathode degradation as a function of temperature, morphology and particle size.

Introduction

Understanding the degradation mechanisms of battery materials like lithium cobalt oxide ($Li_xCoO_2$, LCO) during electrochemical cycling is a multi-scale challenge. At the atomistic scale, the stability of specific lithium arrangements determines phase transitions. At the continuum scale, these transitions manifest as microstructural evolutions—patterns of lithium-rich and lithium-poor regions—that induce mechanical strain and eventual particle failure. Historically, bridging these scales has been difficult due to the lack of thermodynamically consistent free energy functions for continuum models like phase field theories.

Methodology: A Scale-Bridging Framework

This study introduces a novel pipeline that bridges the gap between first-principles electronic structure calculations and macroscopic material behavior using "integrable" deep neural networks (IDNNs).

The multi-scale modeling framework bridging atomic-scale DFT and Monte Carlo simulations to continuum-scale phase field modeling via integrable deep neural networks.

The framework consists of five primary stages:

  1. Density Functional Theory (DFT): High-accuracy quantum mechanical calculations determine the formation energies of 333 unique $Li_xCoO_2$ configurations.
  2. Cluster Expansion (CE): A surrogate model is trained on DFT data to efficiently map the energy landscape of the crystal lattice.
  3. Monte Carlo (MC) Sampling: The CE model enables large-scale simulations of thermodynamic behavior at various temperatures, generating chemical potential data.
  4. Integrable Deep Neural Networks (IDNNs): The core innovation involves using IDNNs to represent the derivative of the free energy. By analytically integrating the network, the authors obtain a smooth, thermodynamically consistent free energy function required for phase field modeling.
  5. Phase Field Modeling: The physics-informed free energy is incorporated into continuum equations to simulate the microstructural evolution of battery particles.

Results and Discussion

Calibration and Voltage Prediction

To ensure physical accuracy, the model was calibrated against experimental voltage-composition curves. Specifically, the "Hubbard U" parameter, which corrects for electronic correlation in DFT, was optimized to match experimental benchmarks.

Comparison of predicted lithiation voltages from DFT (using the optB88 functional) against experimental data across different composition ranges as a function of the Hubbard U parameter.

The framework successfully predicted average lithiation voltages that closely align with experimental observations, particularly in the $x \in [0.5, 1]$ range.

Phase Stability and Microstructure

The simulations revealed that LCO's internal structure is highly sensitive to temperature. At 340 K, lithium distribution remains relatively uniform. However, at lower temperatures (260 K and 300 K), the material undergoes spinodal decomposition, leading to phase separation. At 300 K, ordered phases appear as circular precipitates, while at 260 K, they become elongated and tortuous, likely increasing internal mechanical strain.

Impact of Particle Size

A significant finding of this work is the theoretical explanation for the benefits of nano-sizing. Comparing 1 $\mu$m particles to 50 nm particles, the simulations showed that nanoparticles reach equilibrium 100 to 10,000 times faster. Furthermore, smaller particles maintain a more homogeneous lithium distribution, reducing the risk of degradation compared to larger particles.

Conclusion and Future Outlook

The IDNN-enabled scale-bridging framework provides a rigorous method for incorporating atomistic physics into continuum-level simulations. While the current model focuses on the O3 host structure, future work will expand to include the H1-3 phase transition and integrate mechanical strain effects and interphase interface energies. This approach paves the way for more accurate predictions of battery life and the design of more resilient cathode materials.

Summary updated

Abstract

Li$_xTM$O$_2$ (TM={Ni, Co, Mn}) are promising cathodes for Li-ion batteries, whose electrochemical cycling performance is strongly governed by crystal structure and phase stability as a function of Li content at the atomistic scale. Here, we use Li$_x$CoO$_2$ (LCO) as a model system to benchmark a scale-bridging framework that combines density functional theory (DFT) calculations at the atomistic scale with phase field modeling at the continuum scale to understand the impact of phase stability on microstructure evolution. This scale bridging is accomplished by incorporating traditional statistical mechanics methods with integrable deep neural networks, which allows formation energies for specific atomic configurations to be coarse-grained and incorporated in a neural network description of the free energy of the material. The resulting realistic free energy functions enable atomistically informed phase-field simulations. These computational results allow us to make connections to experimental work on LCO cathode degradation as a function of temperature, morphology and particle size.
BibTeX
@misc{teichert2021li,
  author = {Gregory H. Teichert and Sambit Das and Muratahan Aykol and Chirranjeevi Balaji Gopal and Vikram Gavini and Krishna Garikipati},
  title = {{Li$\_x$CoO$\_2$ phase stability studied by machine learning-enabled scale bridging between electronic structure, statistical mechanics and phase field theories}},
  year = {2021},
  eprint = {2104.08318},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2104.08318},
}
Figure