Modeling phase transformations in Mn-rich disordered rocksalt cathodes with machine learning interatomic potentials

This study employs machine learning interatomic potentials to reveal that Mn-rich disordered rocksalt cathodes undergo a phase transformation into a spinel-like structure driven by transition metal migration rather than Mn$^{2+}$ formation, resulting in improved lithium transport kinetics and higher capacity.

The Gist

Imagine a battery as a bustling city where tiny lithium ions are the commuters, and the battery's cathode is a massive, crowded apartment building. For years, scientists have been trying to build better buildings for these commuters. One promising design is called a "Disordered Rocksalt" (DRX) building. It's like a chaotic apartment complex where the residents (manganese, titanium, and other atoms) are thrown in randomly, with no specific rules about who lives where.

The problem is that in this chaotic building, the lithium commuters sometimes get stuck, making the battery slow and less powerful. However, recent experiments showed something magical happens: after the battery is used a few times (charged and discharged), this chaotic building spontaneously rearranges itself into a more organized, "spinel-like" structure. This new structure allows lithium to move much faster, boosting the battery's performance.

The big question was: How does this messy building magically clean itself up, and what exactly is happening inside?

This is where the researchers, led by Peichen Zhong and Gerbrand Ceder, stepped in. They couldn't watch this happen in real life because it happens too fast and at a scale too small for human eyes. Instead, they built a super-smart digital twin of this building using a type of artificial intelligence called a "Machine Learning Interatomic Potential" (MLIP).

Here is a simple breakdown of what they discovered:

1. The "Smart Architect" (The AI Model)

Traditional computer simulations are like trying to calculate the weight of every single brick in a building by hand—it takes forever and is too slow to see the whole picture. The researchers used a pre-trained AI (called CHGNet) that had already learned the basic laws of physics for many materials. They then "fine-tuned" this AI specifically for their manganese-rich battery material.

Think of this AI as a super-architect who can predict exactly how every atom will move and react, but it does it millions of times faster than traditional methods. This allowed them to run a simulation that lasted for a "nanosecond" (a billionth of a second), which is an eternity in the world of atoms.

2. The Great Rearrangement (Phase Transformation)

They started their simulation with the chaotic, disordered building. As they watched the "movie" of the atoms moving:

• The Migration: The manganese atoms (the heavy furniture in our apartment analogy) started to shuffle around. They moved from their random spots into specific, organized rows.
• The Trigger: A common theory was that these atoms only moved because they changed their electrical charge (like a person changing their mood). However, the AI simulation revealed a twist: The atoms started moving before they fully changed their charge.
• The Result: The manganese atoms organized themselves into a specific pattern (the "spinel-like" or δ-phase). Once this pattern was established, the atoms settled into a new, lower-energy state. It's like a messy room suddenly snapping into a perfect, organized layout because the furniture found a more comfortable fit.

3. The "Highway" Effect (Why it's Better)

The most important discovery was about the "roads" inside the building.

• In the messy building, the lithium commuters had to navigate through narrow, blocked paths.
• In the new, organized building, the manganese atoms moved aside to create wide, open highways (called "0-TM channels") where only lithium and empty space existed.
• The Analogy: Imagine a crowded hallway where people are blocking the way. If the people move to the sides and form a neat line, a clear path opens up for the emergency responders (lithium ions) to zoom through. This is why the battery becomes faster and holds more energy.

4. The Charge Mystery

The researchers also looked at the "mood" (valence state) of the manganese atoms. They found that while some manganese atoms did change their charge (becoming "Mn2+"), this happened after the structure had already started organizing.

• Old Theory: The atoms changed their mood first, which forced them to move.
• New Finding: The atoms moved first to organize the building, and then their moods changed to match the new order. The organization caused the charge change, not the other way around.

5. The Battery's Performance

Finally, they simulated how the battery would behave electrically.

• The Old Messy Building: When you tried to charge it, the voltage (the "pressure" pushing the lithium) would jump up and down erratically, like a bumpy ride.
• The New Organized Building: The voltage became smooth and steady, like a highway cruise.
• The Capacity: The new structure could hold more lithium than the original messy one, and it could do so without the structural stress that usually breaks batteries down over time.

Summary

In short, this paper used a super-fast AI to watch a chaotic battery material reorganize itself into a highly efficient, ordered structure. They discovered that the atoms move to create a better layout first, and the electrical changes follow. This new layout creates "highways" for lithium, making the battery faster, stronger, and more stable. It's a bit like watching a chaotic crowd of people spontaneously form an orderly queue, creating a clear path for everyone to move faster.

Technical Summary

Technical Summary: Modeling Phase Transformations in Mn-rich Disordered Rocksalt Cathodes with Machine Learning Interatomic Potentials

Problem Statement
Mn-rich disordered rocksalt (DRX) cathode materials, such as Li$_x$Mn$_{0.8}$Ti$_{0.1}$O$_{1.9}$F$_{0.1}$, exhibit a beneficial in-situ phase transformation from a disordered structure to a partially disordered spinel-like phase (termed the $\delta$-phase) during electrochemical cycling. This transformation is associated with increased capacity and rate capability. However, the atomic-scale mechanisms driving this transition—specifically the coupled dynamics of transition metal (TM) migration, charge state evolution, and cation ordering—are difficult to model. Conventional ab initio methods are computationally prohibitive for the required time and length scales, while traditional cluster expansion (CE) models lack the expressibility to capture complex charge-coupled dynamical processes and migration barriers in diverse local environments.

Methodology
The authors employed a multi-stage computational workflow centered on a fine-tuned Machine Learning Interatomic Potential (MLIP) based on the CHGNet architecture.

1. Dataset Construction: A comprehensive sampling of the Li–Mn–Ti–O–F chemical space was performed using r2SCAN-DFT calculations. This included:
   • Existing stable compounds from the Materials Project and prior DRX-CE datasets.
   • Custom-generated partially disordered structures with varying degrees of spinel-like ordering, created via Monte Carlo (MC) simulations and site-exchange protocols.
   • Specific sampling of TM migration pathways (octahedral-tetrahedral-octahedral, or o-t-o, and direct o-o paths) at various delithiation levels to capture high-energy barrier events.
2. Model Training: A "Trial-CHGNet" was initially fine-tuned on relaxation trajectories. Subsequently, a "Sample-CHGNet" was parameterized using static DFT calculations of the sampled migration pathways and MD trajectories to enhance sampling of thermally excited configurations. Finally, a "Product-CHGNet" was fine-tuned on a training set of 88,852 structures, achieving test errors of ~2 meV/atom (energy) and ~68 meV/Å (force).
3. Simulation: Charge-informed molecular dynamics (MD) simulations were conducted on a delithiated Li$_{0.6}$Mn$_{0.8}$Ti$_{0.1}$O$_{1.9}$F$_{0.1}$ system at 1273 K. The MLIP utilized predicted magnetic moments as a proxy for atomic charge to infer valence states (Mn$^{2+}$, Mn$^{3+}$, Mn$^{4+}$) dynamically.
4. Electrochemical Analysis: Voltage profiles for the transformed $\delta$-phase, the original DRX phase, and a fully ordered spinel phase were calculated by constructing convex hulls of Li-vacancy configurations and analyzing intercalation reactions.

Key Results

• Phase Transformation Mechanism: The MD simulations successfully captured the transition from a disordered rocksalt to a spinel-like $\delta$-phase. The transformation is thermodynamically favored, evidenced by a ~60 meV/anion energy reduction.
• Migration and Ordering Sequence: The study reveals a specific sequence of events:
  • Mn migration initiates primarily as Mn$^{3+}$ moving through tetrahedral sites, occurring before the establishment of long-range spinel-like ordering.
  • The emergence of tetrahedral Mn$^{2+}$ is a consequence of the spinel-like ordering, rather than the trigger for cation migration.
  • The transformed structure exhibits a higher concentration of non-transition metal (0-TM) face-sharing channels (tetrahedral sites surrounded only by Li or vacancies), which are critical for Li percolation.
• Electrochemical Signatures:
  • Unlike the ordered spinel Li$_x$MnO$_2$, which exhibits a two-phase reaction with a sharp voltage step (>1 V) upon delithiation below $x \approx 0.6$, the $\delta$-phase maintains a solid-solution behavior with a pseudo-plateau at low voltages.
  • The $\delta$-phase demonstrates a higher Li capacity derived from the 0-TM to tetrahedral-Li conversion reaction compared to both the DRX and fully ordered spinel phases.
  • The DRX phase shows restricted capacity around 3 V due to a lack of 0-TM channels and a steeper voltage slope.

Significance and Claims
The paper claims to provide atomic-level insights into the solid-state phase transformation in Mn-rich DRX cathodes and its direct relation to experimental electrochemistry. The authors highlight that their work demonstrates the potential of MLIPs, specifically charge-informed CHGNet, to model complex oxide materials where charge dynamics and structural evolution are tightly coupled. By resolving the timescale discrepancies between charge disproportionation and ion hopping, the study challenges the previous notion that Mn$^{2+}$ formation triggers migration, suggesting instead that it follows the establishment of spinel-like ordering. The work offers a mechanistic understanding of how partial cation disorder eliminates detrimental two-phase reactions while enhancing Li transport kinetics and capacity, validating the utility of MLIPs for studying complex energy materials.