Hierarchical high-throughput screening of alkaline-stable lithium-ion conductors combining machine learning and first-principles calculations

This study employs a hierarchical high-throughput screening workflow combining machine learning and first-principles calculations to identify over 200 alkaline-stable NASICON and garnet-type lithium-ion conductors, revealing key cation substitution strategies and design trade-offs for advancing solid-state humid Li-air batteries.

The Gist

Imagine you are trying to build a super-efficient, long-lasting battery for a future where we power our world with clean energy. Specifically, you want to build a Solid-State Lithium-Air Battery. Think of this as a "Holy Grail" battery: it's light, holds a massive amount of energy, and could power electric planes or cross-country trucks.

However, there's a catch. To make these batteries work well, they need to breathe in the air. But real air isn't just oxygen; it's full of moisture (humidity). When this battery breathes in humid air, it creates a byproduct called Lithium Hydroxide (LiOH).

Here is the problem: LiOH is like a super-strong acid (but on the basic side, like drain cleaner). It creates a highly corrosive, alkaline environment that eats away at the battery's internal components. Most of the materials we currently use to conduct electricity inside these batteries dissolve or break apart when they touch this "drain cleaner" environment.

This paper is a massive digital treasure hunt to find the "superheroes" of battery materials that can survive this harsh, wet, alkaline environment.

The Detective Work: A Two-Step Screening Process

The researchers didn't just guess; they built a high-tech, automated detective system to sift through 320,000 potential chemical recipes. They used a "hierarchical" approach, which is like a job interview process:

1. The Resume Filter (Machine Learning): First, they used a super-fast AI brain (called CHGNet) to quickly scan all 320,000 candidates. This AI is like a hiring manager who can read thousands of resumes in a second. It quickly eliminated the obvious failures—materials that are too unstable or impossible to make.
2. The Deep Dive (Supercomputer Calculations): The survivors (about 3,000 of them) were sent to a more rigorous, slower, but highly accurate test using Density Functional Theory (DFT). This is like a grueling, multi-day interview where they check every single detail to ensure the candidate is truly qualified.

After this intense screening, they found 209 "golden" candidates that could survive the alkaline attack.

The Two Main Contenders: NASICON vs. Garnet

The researchers focused on two famous families of battery materials, which we can think of as two different types of castles:

• The NASICON Castle: These are like castles built with phosphate bricks.

  • The Weakness: The mortar holding the bricks together (the phosphate groups) dissolves easily in the alkaline "drain cleaner."
  • The Fix: The researchers found that swapping some of the bricks for early transition metals (like Scandium, Titanium, or Zirconium) acts like reinforcing the castle walls with steel. These metals form a tough, protective skin (passivation layer) that stops the castle from crumbling.
  • The Twist: They also found that mixing in some silicate bricks makes the castle even more resistant to the acid, though it makes the castle harder to build in the first place.

• The Garnet Castle: These are like castles built with lanthanide gems (specifically Lanthanum).

  • The Strength: These castles are naturally tougher against the alkaline attack because they don't have the weak phosphate mortar.
  • The Secret Weapon: Even if the castle starts to crumble, the Lanthanum gems inside naturally form a protective shield (a passivation layer) that stops the damage from spreading. It's like having a self-healing armor.
  • The Best Materials: The study found that swapping in elements like Tungsten, Tantalum, and Niobium made these Garnet castles incredibly stable.

The Great Trade-Off: Speed vs. Safety

Here is the tricky part, the "Catch-22" of battery design:

• The Goal: You want the battery to conduct electricity fast (high conductivity) so your phone charges in seconds.
• The Problem: To make the battery conduct faster, you usually need to stuff more Lithium ions into the material.
• The Conflict:
  • In NASICON castles, you can add more Lithium to make it faster without making it weaker against the acid. It's a win-win!
  • In Garnet castles, adding more Lithium to make it faster actually makes the walls weaker against the acid. It's a zero-sum game. If you want speed, you lose safety. If you want safety, you lose speed.

The "Mixed" Heroes (MIECs)

The researchers also looked for materials that can conduct both electricity (ions) and electrons. These are called Mixed Ionic-Electronic Conductors (MIECs). Think of them as "dual-language" materials that can do double duty.

However, to conduct electrons, you usually need "redox-active" metals (metals that can easily change their charge). The problem is, these metals tend to dissolve in the alkaline soup. The researchers found a few rare combinations (like adding Iron) that might work, but it's a very delicate balance. It's like trying to build a bridge out of rubber and steel; you need just the right mix so it doesn't snap or melt.

The Big Takeaway

This paper is a roadmap for the future of batteries. It tells us:

1. We can't just use the old materials. They will dissolve in the humid air batteries need to work.
2. We have found 209 new recipes that are chemically stable enough to survive.
3. Design is a balancing act. We have to carefully choose which atoms to swap in to get the perfect mix of:
   • Safety: Not dissolving in the alkaline rain.
   • Speed: Conducting electricity fast.
   • Buildability: Being able to actually manufacture the material in a factory.

In short, the researchers used AI and supercomputers to find the "magic ingredients" that will allow us to build batteries that can breathe in humid air without falling apart, paving the way for the next generation of clean energy technology.

Technical Summary

1. Problem Statement

Solid-state batteries (SSBs), particularly next-generation solid-state Li–O₂ batteries, require lithium-ion conductors (Solid-State Electrolytes, SSEs) that possess high ionic conductivity and stability under harsh operating conditions.

• The Specific Challenge: When Li–O₂ batteries operate in humid air, the discharge product is Lithium Hydroxide (LiOH) rather than Lithium Peroxide (Li₂O₂). LiOH is highly hydrophilic and creates a highly alkaline environment (pH ~12–15) on the cathode surface.
• Current Limitations:
  • Sulfide/Halide electrolytes: React spontaneously with water to form toxic gases or degrade rapidly.
  • Oxide electrolytes (NASICON and Garnet): While more water-stable than sulfides, they suffer from alkaline instability.
    • NASICON: The polyanion groups (e.g., phosphate) dissolve in high pH, leading to bulk degradation.
    • Garnet: While resistant to LiOH, they are susceptible to proton exchange (Li⁺/H⁺) and surface degradation in humid environments.
• Goal: Identify new chemical compositions within NASICON and Garnet frameworks that simultaneously offer high alkaline stability, synthesizability, and ionic/electronic conductivity, overcoming the trade-offs inherent in current materials like LiTi₂(PO₄)₃ (LTPO) and Li₇La₃Zr₂O₁₂ (LLZO).

2. Methodology

The authors employed a hierarchical high-throughput screening workflow combining Machine Learning Interatomic Potentials (MLIP) and Density Functional Theory (DFT) to evaluate over 320,000 potential compositions.

A. Chemical Space Definition

• Frameworks: NASICON ($Li_xM_2(AO_4)_3$) and Garnet ($Li_xLa_3M_2O_{12}$).
• Substitutions: Systematic substitution of cations at octahedral (M) and tetrahedral (A) sites with transition metals, post-transition metals, and lanthanides.
• Scope: 313,217 NASICON and 6,823 Garnet candidates were generated by varying Li content ($x$) and cation valence states.

B. Screening Workflow

1. Pre-screening (MLIP):
   • Used the CHGNet universal machine-learning potential (pretrained on Materials Project data) to relax structures and calculate energies.
   • Filtered candidates with energy above the convex hull ($E_{hull}$) < 100 meV/atom.
   • Rapidly evaluated chemical, electrochemical, and alkaline stability metrics.
2. Down-selection (DFT):
   • Re-relaxed the ~3,000 surviving candidates using DFT (PBE functional) for higher accuracy.
   • Applied a stricter $E_{hull}$ cutoff (< 25 meV/atom).
   • Recalculated stability metrics using DFT energies.
3. Conductivity Evaluation:
   • Performed Molecular Dynamics (MD) simulations using fine-tuned CHGNet models to extrapolate Li-ion conductivity at 300 K for the final candidates.

C. Stability Metrics

• Synthesizability: Energy above the convex hull ($E_{hull}$).
• Electrochemical Stability: Oxidation ($V_{ox}$) and reduction ($V_{red}$) limits.
• Chemical Stability: Reaction energy ($\Delta E_{rxn}$) with H₂O and LiOH.
• Alkaline Stability (Key Metric):
  • Grand Pourbaix Potential ($\phi_{pbx}$): A modified Pourbaix diagram accounting for Li⁺ exchange, pH, and voltage.
  • $\Delta \phi_{pbx}$: The decomposition driving force. Lower values indicate better stability.
  • Passivation Index (PI): The fraction of the operating window where solid decomposition products form (indicating a protective passivation layer). High PI (>0.8) is desirable.

3. Key Contributions

1. Discovery of 209 Alkaline-Stable Candidates: Identified 81 NASICON and 128 Garnet compounds that are thermodynamically metastable but likely synthesizable and resistant to alkaline degradation.
2. Mechanistic Insight into Alkaline Stability:
   • NASICON: Stability is limited by the dissolution of polyanion groups (e.g., PO₄³⁻). Stability is improved by substituting early transition metals (Sc, Hf, Zr, Ti) which form stable oxide passivation layers, and by using silicate groups (though these are harder to synthesize).
   • Garnet: Stability is driven by the presence of Lanthanum (La). Even if the octahedral dopants (e.g., W, Ta, Nb) have low passivation potential individually, the high PI of La ensures the formation of a protective La-based ternary oxide passivation layer (e.g., La₂WO₆) upon decomposition.
3. Identification of Fundamental Trade-offs:
   • Garnet: There is an intrinsic trade-off between Li-ion conductivity and alkaline stability. High conductivity requires high Li content (Li-stuffing, $x \approx 7$), but this increases the decomposition driving force ($\Delta \phi_{pbx}$) and promotes proton exchange. The most alkaline-stable garnets have low Li content ($x=3$) and negligible conductivity.
   • NASICON: Conductivity and alkaline stability can be optimized simultaneously by tuning Li content (optimal at $x=1.5$) without a severe trade-off.
4. Mixed Ionic-Electronic Conductors (MIECs): Identified specific redox-active substitutions (e.g., Fe, Co, Mn) that could enable MIEC cathodes, though these generally suffer from lower alkaline stability compared to pure SSEs.

4. Key Results

• Performance vs. Prototypes:
  • NASICON: The best candidate, Li₃Sc₂(PO₄)₃, shows a 48% reduction in $\Delta \phi_{pbx}$ compared to the baseline LTPO.
  • Garnet: The best candidate, Li₃La₃W₂O₁₂, shows a 66% reduction in $\Delta \phi_{pbx}$ compared to the baseline LLZO.
• Elemental Trends:
  • NASICON: Early transition metals (Sc, Ti, Zr, Hf) are optimal for the M-site. Post-transition metals (Al, Ga, In) and alkaline earth metals (Mg, Ca) generally reduce passivation capability unless paired with high-PI elements.
  • Garnet: W, Ta, and Nb are the most effective dopants for lowering decomposition energy, despite their low individual passivation indices, due to the "shielding" effect of La.
• Conductivity:
  • Several candidates (e.g., Li₆.₅La₃Zr₁.₅Ta₀.₅O₁₂) show predicted conductivities > 10 mS/cm at 300 K.
  • MIEC candidates (e.g., LiₓHf₁.₅Fe₀.₅(PO₄)₃) show high conductivity but require careful balancing of redox-active content to maintain alkaline stability.

5. Significance

• Enabling Humid Li–O₂ Batteries: This work provides a concrete materials roadmap for developing solid-state electrolytes capable of operating in the highly corrosive, high-pH environments generated by Li–O₂ batteries in humid air.
• Design Principles: It establishes that surface passivation is the key to long-term stability for oxide electrolytes in alkaline media, rather than absolute thermodynamic stability.
• Methodological Advancement: Demonstrates the efficacy of combining universal ML potentials (CHGNet) with DFT for exploring vast chemical spaces, significantly accelerating the discovery of complex multi-component materials.
• Guidance for Synthesis: By highlighting specific cation substitutions (e.g., Sc in NASICON, W/Ta in Garnet) and identifying the trade-offs between Li-content, conductivity, and stability, the study offers actionable guidelines for experimentalists to synthesize the next generation of solid-state batteries.