Unveiling Mechanisms of SEI Formation and Sodium Loss in Sodium Batteries via Interface Reactor Sampling

This paper introduces an "Interface Reactor" sampling strategy to construct a charge-aware neuroevolution potential, enabling stable, long-timescale simulations that reveal distinct SEI formation mechanisms and sodium loss dynamics in carbonate versus ether-based electrolytes, thereby providing a robust computational framework for rational SEI engineering in sodium-metal batteries.

The Gist

The Big Picture: Why Sodium Batteries Need a "Bodyguard"

Imagine you are building a new type of battery using Sodium (the common table salt cousin of Lithium) instead of the expensive Lithium found in your phone. Sodium is cheap and everywhere, which is great. But there's a problem: Sodium is a bit "messy." When it tries to charge and discharge, it tends to react violently with the liquid inside the battery, creating a chaotic mess that kills the battery's life.

To fix this, scientists need to build a protective shield on the surface of the sodium. In the battery world, this shield is called the SEI (Solid Electrolyte Interphase). Think of the SEI as a bodyguard for the sodium metal.

• A good bodyguard lets the sodium ions (the energy carriers) pass through smoothly but stops the liquid from attacking the sodium.
• A bad bodyguard is either too weak (letting the liquid destroy the sodium) or too sticky (trapping the sodium so it can't move).

The problem is, we didn't know how this bodyguard forms or why some electrolytes (the battery liquids) make a great bodyguard while others make a terrible one. Watching it happen in real life is like trying to film a ghost with a camera that only takes photos once a year. It's too fast and too small to see.

The Breakthrough: The "Time-Traveling Microscope"

The researchers in this paper invented a new way to watch this process happen. They couldn't use a real microscope, so they built a super-smart computer simulation.

Usually, computer simulations of atoms are like trying to run a marathon while carrying a heavy backpack. They are either:

1. Too slow: Accurate but only last for a tiny fraction of a second (picoseconds).
2. Too fast: Can run for a long time but are so inaccurate they get the physics wrong.

The Solution: The "Interface Reactor"
The team created a new AI model (called qNEP) that acts like a high-speed, time-traveling microscope.

• The Analogy: Imagine you are trying to learn a complex dance routine.
  • Old way: You watch a slow-motion video of one step, then guess the next. You often trip and fall (simulation crashes).
  • New way (This paper): You have a dance instructor (the AI) who has seen every possible move. It guides you through the whole routine without you ever tripping, allowing you to watch the entire dance unfold over 100 nanoseconds (which is a lifetime in the atomic world).

This allowed them to watch the "bodyguard" (SEI) form from start to finish, something no one had done accurately before.

The Discovery: Two Different "Bodyguards"

They tested two common types of battery liquids: Carbonate-based (like EC) and Ether-based (like DME). They found these two liquids build completely different kinds of bodyguards.

1. The Carbonate Liquid: The "Chaotic Construction Site"

• What happens: When sodium touches this liquid, it explodes into a reaction. It's like a construction crew throwing bricks, wood, and plastic everywhere at once.
• The Result: The bodyguard (SEI) becomes a messy mix of organic and inorganic junk. It's full of holes and weak spots.
• The Consequence: Because the shield is messy, the liquid keeps attacking the sodium. The battery "leaks" energy, and the sodium gets trapped in the mess, unable to do its job. The battery dies quickly.

2. The Ether Liquid: The "Precision Bricklayer"

• What happens: This liquid is calmer. Instead of a chaotic explosion, it focuses on building one specific, strong material: Sodium Fluoride (NaF).
• The Result: The bodyguard forms a perfect, dense, crystal wall. It's like a high-tech force field.
• The Consequence: This wall is so good that it stops the liquid from attacking the sodium. The sodium can move in and out freely. The battery stays healthy and lasts a long time.

The Key Insight: The researchers realized that the speed of the reaction determines the quality of the shield. Fast, chaotic reactions make bad shields; slow, controlled reactions make perfect shields.

The "Sodium Loss" Mystery

The paper also solved a mystery: Where does the sodium go?
In a bad battery (Carbonate type), you put in 100 units of sodium to charge it, but only 60 come out to power your device. Where did the other 40 go?

• The Trap: In the messy, carbonate shield, the sodium gets "stuck." Imagine trying to walk through a room filled with sticky flypaper. You get stuck, and you can't get out.
• The Freedom: In the Ether shield, the sodium moves through a smooth, open hallway. It gets in and gets out easily.

The researchers used a special simulation technique (Metadynamics) to prove that the chemical nature of the shield decides if the sodium is free or trapped.

Why This Matters

This paper is a huge deal for three reasons:

1. It's a New Tool: They built a "Interface Reactor" that other scientists can use to study any battery, not just sodium ones. It's like giving everyone a new pair of X-ray glasses.
2. It Explains the "Why": We now know why some batteries fail and others succeed. It's all about how the shield forms.
3. It Guides the Future: If we want better batteries, we shouldn't just guess. We should design liquids that encourage the "Precision Bricklayer" (Ether-like) behavior and avoid the "Chaotic Construction Site" (Carbonate-like) behavior.

In short: The researchers used a super-smart AI to watch how a battery builds its own armor. They found that some liquids build a messy, leaky shield, while others build a perfect, impenetrable wall. Now, we know exactly how to build better batteries for the future.

Technical Summary

1. Problem Statement

The performance of sodium-metal batteries (SMBs) is critically limited by interfacial phenomena, specifically the formation and evolution of the Solid Electrolyte Interphase (SEI). While the SEI is essential for suppressing electronic transport while allowing ion conduction, its formation mechanisms and atomic-scale evolution remain poorly understood due to several challenges:

• Experimental Limitations: Techniques like XPS and TEM provide only ex-situ or ensemble-averaged data, failing to capture ultrafast, reactive interfacial processes under operating conditions.
• Computational Limitations:
  • Ab Initio Molecular Dynamics (AIMD): Too computationally expensive for the large system sizes (thousands of atoms) and long timescales (nanoseconds) required to observe SEI maturation.
  • Reactive Force Fields (ReaxFF): Lack transferability and accuracy for complex electrochemical interfaces, often failing to capture single-electron reduction pathways.
  • Standard Machine Learning Potentials (MLPs): While accurate, they often suffer from instability and error accumulation during long-timescale simulations of heterogeneous interfaces, typically limiting simulations to picoseconds.

The core bottleneck is the inability to perform stable, first-principles-accurate molecular dynamics (MD) simulations over hundreds of nanoseconds for systems containing tens of thousands of atoms to capture the kinetic evolution of SEI growth and sodium loss.

2. Methodology: The "Interface Reactor" Strategy

The authors propose a novel computational framework centered on a Charge-Aware Neuroevolution Potential (qNEP) constructed via an "Interface Reactor" sampling strategy.

• qNEP Model: A machine learning potential that explicitly incorporates trainable atomic charges to capture charge-transfer dynamics and long-range electrostatic interactions, which are crucial for electrolyte decomposition and ion migration.
• Interface Reactor Sampling (Active Learning): To ensure the potential remains stable and physically accurate over long timescales, the authors implemented a three-step data generation loop:
  1. High-Temperature Descriptor Sampling: Using high-temperature AIMD to ensure broad coverage of the descriptor space (radial and angular descriptors), enhancing ergodicity.
  2. AIMD-Guided Physical Correction: During active learning, if the MLP-driven trajectory approaches a bifurcation point (risking unphysical configurations), short-range AIMD simulations are triggered to "steer" the system back to physically reasonable reaction pathways. This prevents the accumulation of errors and systematic drift.
  3. Chemical Composition Coverage: The training set explicitly includes intermediates and products from inorganic salts (e.g., NaF, Na₂O) and organic decomposition pathways across bulk, interfacial, and substrate regions.
• Simulation Scale: The trained model enables MD simulations of systems with ~27,000 atoms for up to 100 nanoseconds on a single GPU, achieving a stability improvement of 2–3 orders of magnitude over existing methods.

3. Key Contributions

• Methodological Breakthrough: Established a robust computational framework (Interface Reactor + qNEP) that bridges the gap between microscopic structure and macroscopic performance by enabling stable, DFT-accurate simulations on the nanosecond scale.
• Discovery of Distinct SEI Formation Mechanisms: Revealed fundamentally different growth patterns for carbonate-based vs. ether-based electrolytes.
• Mechanistic Insight into Sodium Loss: Elucidated the atomic-scale origins of irreversible sodium trapping and electrolyte consumption, linking SEI composition to deposition morphology.
• Validation: Correlated simulation results with experimental Gas Chromatography (GC) and X-ray Photoelectron Spectroscopy (XPS) data.

4. Key Results

A. Distinct Decomposition Pathways

• Carbonate Electrolytes (e.g., EC): Exhibit high reactivity. Decomposition follows a "mixed co-formation" mechanism involving rapid two-electron transfers (cleaving C=O bonds) and single-electron transfers. This produces a heterogeneous SEI rich in organic-inorganic matrices (Na₂CO₃, organic sodium salts) and gaseous by-products (ethylene, CO/CO₂).
• Ether Electrolytes (e.g., DME): Exhibit lower reactivity. Decomposition is dominated by the salt (NaPF₆) rather than the solvent. This leads to a "surface-energy-controlled" mechanism where NaF forms rapidly and acts as a dense, self-limiting barrier.

B. SEI Growth Patterns

• Carbonate Systems: The high reactivity of the solvent drives the simultaneous nucleation of various products, resulting in a porous, mixed organic-inorganic SEI.
• Ether Systems: The preferential decomposition of the salt leads to rapid NaF formation. Driven by surface energy minimization, NaF grows laterally in a spherical cap morphology, forming a dense, ordered, inorganic-rich overlayer that kinetically locks the interface and suppresses further solvent decomposition.

C. Regulation of Sodium Storage and Loss

Using Metadynamics simulations, the study analyzed sodium deposition through different SEI layers:

• NaF-Rich SEI (Ether-based): NaF creates a deep potential well that initially traps Na⁺ ions in a "solid solution" state. Once saturation is reached, metallic sodium deposits efficiently with minimal side reactions. This results in high Coulombic efficiency.
• Na₂CO₃-Rich SEI (Carbonate-based): The covalent nature of carbonate ions allows the SEI matrix to adopt a "metallic-like" character easily. This triggers rapid, irreversible electrolyte decomposition. Sodium ions are irreversibly "trapped" within the SEI defects rather than depositing as metal, leading to significant sodium loss and electrolyte depletion.

D. Experimental Corroboration

• GC Analysis: Confirmed the dominance of ethylene (C₂H₄) in EC decomposition and ethane/methane in DME, matching simulation predictions.
• XPS Analysis: Validated the higher content of Na₂CO₃ in EC-derived SEIs and higher NaF/organic sodium content in DME-derived SEIs.

5. Significance

• Rational Design Guidelines: The work provides a clear mechanistic link between solvation structure, interfacial reaction networks, and electrochemical performance. It suggests that engineering electrolytes to promote NaF-rich, inorganic SEIs (via ether solvents or additives) is crucial for improving sodium battery stability.
• General Framework: The "Interface Reactor" sampling strategy is not limited to sodium batteries; it offers a transferable, general approach for simulating complex interfacial reactions in any electrochemical system where long-timescale stability and charge transfer are critical.
• Solving the "Sodium Loss" Mystery: By identifying that irreversible sodium loss is caused by the trapping of ions in covalent SEI matrices (like carbonates) versus the efficient metallic deposition in ionic matrices (like NaF), the study offers a fundamental explanation for low Coulombic efficiency in certain electrolyte systems.