Geometric landscapes for material discovery within energy-structure-function maps.

Area of Science:

• Materials Science
• Crystallography
• Computational Chemistry

Background:

• Porous molecular crystals are a distinct class of porous materials, differing from extended frameworks like metal-organic frameworks (MOFs).
• Energy-structure-function (ESF) maps aid in predicting crystal structures and properties for discovering new porous materials.
• Representing and navigating high-dimensional ESF maps to identify optimal polymorphs remains a significant challenge.

Purpose of the Study:

• To introduce a novel representation, "geometric landscapes," for exploring complex ESF maps of porous molecular crystals.
• To demonstrate how geometric landscapes facilitate the identification of energetically favorable and functionally relevant crystalline phases.
• To establish geometric landscapes as a descriptor for predicting porous material performance in applications like gas adsorption.

Main Methods:

• Utilizing persistent homology to quantify geometric similarity for creating geometric landscapes.
• Developing a machine learning model trained on geometric similarity descriptors.
• Applying the method to analyze ESF maps and predict material properties.

Main Results:

• Geometric landscapes effectively represent high-dimensional ESF maps, enabling easier exploration.
• The method automatically identifies promising crystalline phases within complex structural landscapes.
• A machine learning model using geometric similarity achieved high accuracy in predicting methane storage performance.

Conclusions:

• Geometric landscapes offer a powerful new tool for the discovery and design of porous molecular crystals.
• This approach simplifies the identification of materials with targeted functionalities for applications such as gas storage.
• The developed methodology advances the predictive capabilities in materials science for porous materials.