Superionic Ionic Conductor Discovery via Multiscale Topological Learning.

Area of Science:

• Materials Science
• Computational Chemistry
• Solid-State Physics

Background:

• Lithium superionic conductors (LSICs) are vital for next-generation solid-state batteries, promising high ionic conductivity and safety.
• Discovering new LSICs is hindered by vast chemical spaces, limited data, and complex structure-property relationships for ion transport.
• Existing methods struggle to efficiently navigate the complexities of LSIC material design.

Purpose of the Study:

• To introduce a novel multiscale topological learning (MTL) framework for efficient LSIC discovery.
• To overcome challenges in identifying LSICs by integrating algebraic topology and unsupervised learning.
• To develop a scalable tool for accelerating materials discovery in energy storage.

Main Methods:

• Developed an MTL framework modeling lithium-only and lithium-free substructures to extract topological features.
• Introduced topological screening metrics (cycle density, minimum connectivity distance) for structural connectivity and ion diffusion.
• Employed unsupervised clustering for candidate identification and ab initio molecular dynamics for validation.

Main Results:

• The MTL framework successfully identified 14 novel lithium superionic conductor candidates.
• Four of the discovered LSICs were independently validated through experimental testing.
• The framework demonstrated significant acceleration in the LSIC discovery pipeline.

Conclusions:

• The multiscale topological learning framework provides an efficient and scalable approach to discovering novel LSICs.
• This study validates the power of integrating topological data analysis with machine learning for materials science.
• The developed methodology offers a promising pathway for advancing solid-state battery technology.