Reproduce “Inverse design of two-dimensional materials with invertible neural networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Victor Fung, Jiaxin Zhang, Guoxiang Hu, P. Ganesh, Bobby G. Sumpter
A framework called MatDesINNe, utilizing Invertible Neural Networks and conditional variants, performs inverse design for materials by generating material parameters from desired properties. The framework achieves near-chemical accuracy (MAE "=" 0.1 eV or lower via DFT validation) and over five orders of magnitude speedup in identifying candidates for band gap engineering in 2D MoS2.
Key insights
- The intrinsic invertibility of INNs, combined with post-processing, significantly enhances the stability and accuracy of inverse design compared to traditional generative models.
- Specific design parameters, such as shear strain, exhibit higher sensitivity in tuning the band gap of MoS2, and the framework can map regions for metal-insulator transitions (MITs).
- High-fidelity material candidates can be generated while maintaining diversity, addressing a common limitation of other generative models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.